Beruflich Dokumente
Kultur Dokumente
Statistical Modelling in R
Course Website: http://isites.harvard.edu/icb/icb.do?keyword=k91599
Aedin Culhane
Email: aedin@jimmy.harvard.edu
http://www.hsph.harvard.edu/research/aedin-culhane/
http://bcb.dfci.harvard.edu/aedin/courses
Contents
i
2.6.2 Installing packages using RStudio . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.6.3 Commands to install packages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.6.4 Package help . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.6.5 Load package into workspace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.7 Customizing R Start-up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.7.1 R Studio Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.7.2 Rprofile . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3 Objects in R 28
3.1 Using ls and rm to managing R Objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.2 Types of R objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.2.1 vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2.2 Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2.3 list . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.2.4 data.frame . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.2.5 factor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.2.6 ordered factor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.2.7 array . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.3 Attributes of R Objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.3.1 Dimension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.4 Creating and accessing objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.5 Modifying elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.5.1 Sorting and Ordering items . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.5.2 Sort a matrix by 2 columns,one in decreasing order, the second ascending . . . . . . 40
3.5.3 Creating empty vectors and matrices . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.5.4 Missing Values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.6 Quick recap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.7 Exercise 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
ii
4.8.1 Reading data from SAS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.8.2 S . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.8.3 Stata or Systat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.9 From a Database Connection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.10 Sampling and Creating simulated data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.11 Exercise 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
6 Introduction to graphics in R 91
6.1 The R function plot() . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
6.2 Exercise 7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
6.2.1 Arguments to plot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
6.2.2 Other useful basic graphics functions . . . . . . . . . . . . . . . . . . . . . . . . . 97
6.3 Customize plot with low-level plotting commands . . . . . . . . . . . . . . . . . . . . . . . 110
6.4 Default parameters - par . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
6.4.1 Interactive plots in R Studio - Effect of changing par . . . . . . . . . . . . . . . . . 114
iii
6.4.2 R Colors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
6.4.3 More Colors Palettes; RColorBrewer . . . . . . . . . . . . . . . . . . . . . . . . . 118
6.5 Interacting with graphics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
6.5.1 Exercise 8 - Plotting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
6.6 Saving plots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
6.6.1 Rstudio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
6.6.2 Devices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
6.6.3 Difference between vector and pixel images . . . . . . . . . . . . . . . . . . . . . . 125
6.7 Useful Graphics Resources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
iv
8.9.2 Kaplan-Meier curve estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
8.9.3 Cox proportional hazards model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
8.10 Exercise 11: Survival Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
v
R set up script for this manual
We will run this course with R>2.15 and RStudio 0.97.
To ensure you have all of the packages needed to run this course, either
pkgs<-list("lme4","xtable","xlsx","RODBC","Hmisc",
"SAScii","R2HTML","BiocInstaller","plyr",
"RColorBrewer","lattice","ggplot2","googleVis",
"network","igraph","vcd","MASS","survival",
"venneuler", "gplots", "ggmap", "knitr",
"markdown","tm", "wordcloud", "ade4", "gmodels",
"XML", "XLConnect", "scatterplot3d", "manipulate",
"KernSmooth", "foreign")
Biocpkgs<-c("parallel", "annotate")
checkPkg<- function(x) {
print(x)
if (!x%in%installed.packages()[,1])
install.packages(x)
}
checkBioCPkg<- function(x) {
print(x)
require(BiocInstaller)
if (!x%in%installed.packages()[,1])
biocLite(x)
}
lapply(pkgs, checkPkg)
lapply(Biocpkgs,checkBioCPkg)
search()
This script checked to see if you have installed each package on your computer. If it is installed, its loads
the package into your R session. If it is not installed, it downloads the package from CRAN or Bioconductor
and installed it.
The function search lists the packages that are loaded into your current R session. Well talk more about
packages, how to install them and how to load them during the course.
1
Chapter 1
demo()
demo(graphics)
demo(nlm)
Demo will run through each example, hit return to view each, once it is done you will see the command
prompt > again.
Note if a command is not complete on one line (missing close bracket or quote), R will use continua-
tion prompt +
2
To find out more about parameters and options for the function demo, lets look at help for demo. To get
help simply use the command help or type a question mark ? before the command
help(demo)
`?`(demo)
If you wish to exit R, you can click on the red X in the upper corner of the window frame (Windows),
press CMD-q (MacOS) or Ctrl-D (Linux). On all platforms, you can type
q()
It will ask you if you wish to save the workspace image, and gives you three choices y/n/c.
Well talk more about the workspace later, for now just press c but at least you know how to exit should
you wish to.
rnorm(5)
rnorm(5)
rnorm(10)
rnorm(10, mean = 5, sd = 2)
If you wish to generate the same set of random numbers each time, you could set.seed(10)
You can view previous expressions entered into the R session using the function history. By default only
25 expresssion are show, if you wish to see more type the number as an argument to the function history.
history()
history(50)
Browsing, saving and searching is discussed in more detail later in section refsec:history. In Rstudio you
can also view expression history in the history tab on top right panel (see section refsec:RStudioHistory for
further details.
3
1.3 Getting started: R as a big calculator
Type the following into an R session (or copy and paste from this document).
2 + 2
2 * 2
2 * 100/4
2 * 100/4 + 2
2 * 100/(4 + 2)
210
11%%2 #modulus
log(2)
log(2, base = 2)
Each of these are called expressions. When you type an expression into R, R evaluates the code and
prints the results. The result is not stored or saved to a file. In order to manuiplate expressions we need to
assign the result of an expression to a variable.
1.4 Assignment
Even in the simple use of R as a calculator, it is useful to store intermediate results.Lets store the value of
(tmpVal=log(2) in a symbolic variable tmpVal.
## [1] 0.6931
tmpVal
## [1] 0.6931
exp(tmpVal)
## [1] 2
Note when you assign a value to such a variable, there is no immediate visible result. We need to
print(tmpVal) or just type tmpVal in order to see what value was assigned to tmpVal
2 * 52
## [1] 50
4
x <- 2 * 52
print(x)
## [1] 50
Variables can be assigned using the assignment operators: '<-', '=', '->'
'<-' is the most popular assignment operator, and '=' is a recent addition.
When assigning a value spaces are ignored so 'z<-3' is equivalent to 'z <- 3'
## [1] 50
y==z
## [1] TRUE
Note = and == have very different uses in R. == is a binary operator, which test for equality (A==B
determines if A is equal to B ).
not equal: !=
5
1 == 1
## [1] TRUE
x <- 1:10
y <- 10:1
x > y & x > 5
## [1] FALSE FALSE FALSE FALSE FALSE TRUE TRUE TRUE TRUE TRUE
OR |
Returns TRUE where at least one comparison returns TRUE.
x == y | x != y
## [1] TRUE TRUE TRUE TRUE TRUE TRUE TRUE TRUE TRUE TRUE
NOT !
The ! sign returns the negation (opposite) of a logical vector.
!x > y
## [1] TRUE TRUE TRUE TRUE TRUE FALSE FALSE FALSE FALSE FALSE
These return a logical vector of TRUE or FALSE and can be useful for filtering (we will see this later)
1.6.1 R functions
Within R, you can find help on any command (or find commands) using the following:
help or ?
If you know the command type help or ?
6
## help or ? will open a help page
help(lm)
# ?matrix
`?`(matrix)
apropos
If you know part of the function name you can perform a search of all command to retrieve a list of
functions with a partial match. It is not case sensitive
The functions find() or apropos() will perform a search for a commands that partial match the input
argument
apropos("fish")
## [1] "fisher.test"
apropos("Fisher")
## [1] "fisher.test"
Example
Almost all functions have an example, use the command to get an example of how to use the function
This will run all the examples included with the help for a particular function. If we want to run
particular examples, we can highlight the commands in the help window and submit them by typing
CtrlV
help.search or ??
If you dont know the command and want to do a keyword search for it. For example we wished to
find the command to perform a Students t-test, we could perform a search of function help pages
help.search("Student")
help.start()
`?`(`?`(Student))
Note using the double question mark ?? or the command help.search are equivalent.
RSiteSearch
The function RSiteSearch performs a key words search of R-help mailing list archives, help pages,
vignettes or task views at the website at http://search.r-project.org which is maintained
by Jonathan Baron and the School of Arts and Sciences of the University of Pennsylvania.
7
RSiteSearch("Student's t-test")
help.search will open a html web browser or a MSWindows help browser (depending on the your
preferences) in which you can browse and search R documentation.
In addition, there numerous useful resources for learning R on the web including the R project http:
//www.r-project.org and its mailing lists but also I recommend the following:
In the December 2009 issue of the R Journal. Transitioning to R: Replicating SAS, STATA, and SU-
DAAN Analysis Techniques in Health Policy Data. Anthony Damico http://journal.r-project.
org/archive/2009-2/RJournal_2009-2_Damico.pdf
SAS and R. A blog devoted to examples of tasks (and their code) replicated in SAS and R http:
//sas-and-r.blogspot.com/
R for SAS and SPSS Users. Download a free 80 page document, http://rforsasandspssusers.
com/
R for SAS and SPSS Users which contains over 30 programs written in all three languages.
8
1.7 Quiz
Find a function that will calculate the mean and standard deviation.
Get help on the function rnorm. What is the default mean and standard deviation of the random
normal distribution from which numbers are sampled?
Draw 10, 100, 1000 number from a random normal distributionand assign these to the objects r10,
r100 and r1000 respectively.
z <- 1:11
Z <- 2:22
z == Z
Q2
a <- LETTERS[1:5]
b <- c("a", "b", "c", "d", "e")
a == b
a[1] %in% b # x%in%y tests if x is a member of y
Q3
a <- 1:5
b <- a%%2 #%% is modulus
b == TRUE
9
1.8 A few important points on R
R is case sensitive, i.e. myData and Mydata are different names
x
## [1] 1 2 3 4 5 6 7 8 9 10
## [1] 10 9 8 7 6 5 4 3 2 1
Z <- 20
x == z
## [1] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
x == Z
## [1] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
Comments can be put anywhere. To comment text, insert a hashmarh #. Everything following
it to end of the line is commented out (ignored, not evaluated).
Quotes, you can use both double or single quotes, as long as they are matched.
For names, normally all alphanumeric symbols are allowed plus '.' and '_' Start names with
a character [Aa-Zz] not a numeric character [0-9]. Avoid using single characters or function
names t, c, q, diff, mean, plot etc.
Arguments (parameters) to a function calls f(x), PROC are enclosed in round brackets. Even if
no arguments are passed to a function, the round brackets are required.
print(x)
getwd()
Commands can be grouped together with braces ('{' and '}'). They can be separate by semi-
colon ; or on different lines
{
a <- 1
b <- 3
c <- "What the connection between the Students t-test and Guinness?"
print(a + b + 2)
print(c(a, b) + 2)
print(length(c))
print(nchar(c))
}
## [1] 6
## [1] 3 5
10
## [1] 1
## [1] 61
Bracket Use
() To set priorities 3*(2+4). Function calls f(x)
[] Indexing in vectors, matrices, data frames
{} Creating new functions. Grouping commands {mean(x); var(x)}
[[]] Indexing of lists
Missing values called represented by NA. For more on missing values see section 3.5.4.
Inf and -Inf are positive and negative infite values respectively
1/0
## [1] Inf
-1/0
## [1] -Inf
Sometimes a computation will create a value that doesnt make sense returning an NaN value.
NaN is Not A Number
0/0
## [1] NaN
When you begin your first project in R, you will need to know how to organize your R files and save
your R scripts and output.
The first thing to do when starting an R session, is to ensure that you will be able to find your data and
also that your output will be saved to a useful location on your computer hard-drive. Therefore, set a
working directory
Within R Studio, you can define a RStudio project folder is a great way to start a new project. It aids
with managing R code, data and results.
However even if you dont use the project option or are accessing R through the default or another
interface, you still will need to be able to set a current working directory (in which all your output
files are saved).
11
When I create a new working directory for an R project, I try to include a project Name and Date in
the folder name eg colonJan13.
Within this folder I incrementally name scripts for example S001prepData.R, S0002transform.R,
S003analysis.R, S004results.R etc. I also create a simple file which gives a brief description of each.
We have a lot more tips on scriting in R in section 5.12
There are numerous way to define the working directory in R. To change directory:
1. In the classic R interface. Use the file menu, to change directory File > Change dir
2. If you start R by clicking on an R icon. You may wish to change the default start location by right
mouse clicking on the R icon on the desktop/start menu and changing the Start In property.
For example make a folder C:/work, and make this as a Start in folder
3. In RStudio Tools > Set Working Directory
4. Or in RStudio click on the Files tab (on the bottom right panel). Use the File browser window
to view the contents of a directory and navigate to the directory you wish to set as your home
directory.
click on the triple dot icon on the top right. Navigate to the correct directory
Once you are in the correct directory and see your data files click on the More (blue cog-
wheel), and select Set as Working Directory
To see folders or files in the working directory, use the command dir() (or browse the files using
the Files Browser panel in RStudio)
dir()
dir(pattern = ".txt")
Important side note: R doesnt like windows a back slash (\) that separate folders in a file path.
Indeed it will return a rather cryptic error
12
> setwd("C:\Users\aedin\Documents\Rwork")
Error: "\U" used without hex digits
in character string starting "C:\U"
There are a couple of ways to prevent this, either replace backslash (\) with forward (/) slash or
double back slash (\\) The first back slash tells R to treat the character literally, it is called an escape
character which invokes an alternative interpretation on subsequent characters.
setwd("C:/Users/aedin/Documents/Rwork")
setwd("C:\\Users\\aedin\\Documents\\Rwork")
Using all those slashes can be rather tedious. A nicer way to make scripts work across platform is to
define directories in a more generic way.
The R command file.path will automatically add in the correct slashes for windows, mac or Linux
## [1] "/Documents/Rwork"
Here the tilde symbol ( ) which is the shortcut to your home drive (on any operating system). These
is more details about using these shortcuts in chapter 5.8.1
If you are setting a working directory within a script, you may need to check a folder exisits or create
a new one. These can be achieved with the command file.exists and dir.create respectively
To create a full directory or more complex directory path. A path can be relative to the current location,
in this case two dots mean the directory above
setwd("../../RWork/colonJan13")
Or you can specific a full directory path. For cross-platform compatibility, its best to use file.path()
to create paths. for example to set move the working directory to a folder called Project1 within the
current directory;
13
# If directory doesn't exist create it
if (!file.exists(newdirPath)) dir.create(newdirPath)
setwd(newdirPath)
14
1.9.3 Saving, Loading R workspace and history
1. Saving a R objects
One can either save one or more R object in a list to a file using save() or save the entire R
session (workspace) using save.image().
save(women, file = "women.RData")
save.image(file = "entireL2session.RData")
To load this into R, start a new R session and use the load()
rm(women)
ls(pattern = "women")
load("women.RData")
ls(pattern = "women")
2. Saving R history
R records the commands history of each R session. To view most recent R commands in a
session
history()
help(history)
history(100)
3. Note you can also browse and search history in RStudio, really easily in the history window.
A really nice feature of this window is the ease of sending command either to the console (to
execute code again) or to Source (to a text file or script you are writing in the editor)
15
Chapter 2
After downloading R, click on the R icon. It will open the basic R interface. The windows or macOS
the R interface will pretty similar. The linux interface had fewer options and can be initaed using the
command
verbatim R -g Tk
16
Figure 2.1: R interface
R interface with windows installation of R
Windows
Help - An essential resource!
The icons below will allow one to
open script (.R file),
load image (previous R session, .RData file)
save .R script
copy and paste
stop a computation (this can be an important button, but the ESC also works!)
print.
17
2.2.2 Default R Editor
Within R, one can open an editor using the menu command File -> New script
Can type commands in the editor
Highlight the commands and type CtrlR to submit the commands for R evaluation
Evaluation of the commands can be stopped by pressing the Esc key or the Stop button
Saving (CtrlS) and opening saved commands (CtrlO)
Can open a new editor by typing CtrlN
2.2.3 R Shortcuts
Keyboard Shortcuts for traditional R GUI
Ctrl + A: gets you to the beginning of the line
Ctrl + E: gets you to the end
Ctrl + K: wipes everything to the right of your cursor
Esc: kills the current comment and takes you back to
Up Arrow: brings back last command you typed
Down Arrow: brings back the next command (Up and down basically scrolls up and down through
your history)
Keyboard shortcuts for RStudio are listed online at http://www.rstudio.org/docs/using/
keyboard_shortcuts.
18
2.3 RStudio Interface
RStudio is a free and open source integrated development environment for R. Those familiar with MatLab
will recognize the layout as its pretty similar. RStudio have a brief 2 minute guide to the interface on their
website http://rstudio.org/ which I recommend watching.
On start-up R Studio brings up a window with 3 or 4 panels. If you only see 3 panels, click on File ->
New -> New R Script.
The first thing to notice, is that the bottom left panel console is the exact same as the standard R
console. RStudio just loads your local version of R. You can specify a different version of R (if you have
multiple versions of R running on your machine) by clicking on Tools -> Options and selecting R
version.
2.3.1 R Console
RStudio has a nice console features
start typing a command, for example fi, press the TAB key, it will suggest function that begin with
fi
Select fis for fisher.test. Type
fisher.test(
19
Figure 2.3: Pressing Tab to reveal function help
Tab not only auto-completes, but also suggests parameters and input to the function. Note it says press F1
for further help on this function
and then press the TAB key. You will notice it bring up help on each parameter, you can browse these
and it will help you get familiar with R.
Press F1, it will bring up a help document about the function in the help panel (right bottom)
Press F2, it will show the source code for function
There are many useful keyboard shortcuts in RStudio for a full list of these see http://rstudio.
org/docs/using/keyboard_shortcuts
2.3.2 R Editor
The top left panel is an editor which can be used to edit R scripts (.R), plain text (.txt), html web files or
Sweave (rnw) or markdown (md), the latter two of these can be converted to pdf files. There are several nice
features to this text editor which we will describe during the course. But for now note, that it highlights R
code, and that the code is searchable (click Control-F to search)
In the menu code, you can set preferences to highlight, indent or edit code.
20
Files is a file browser, which allows you to create a new folder, rename a folder or delete a folder.
Click on the triple dot icon (...) on the far right on the menu to browse folders. Under the More menu
you can set your current working directory (more about that below). If you double click on a text, .R,
Sweave or html file it will automatically open in the Editor.
The Plots window displays plots generated in R. Simply type the following command into the Console
window
plot(1:10)
plot(rnorm(10), 1:10)
It will create 2 plots. Use the arrows keys to browse plots, click on zoom, export or delete to manage
plots.
Packages list all of the packages installed in your computer. The packages with tick marked are those
loaded in your current R session. Click on a package name to view help on that package. Note you
can install packages or check for updates. You can also search for a package or search package
descriptions using the search window.
The Help menu provides an extensive R help. The arrows button go forward or back through recent
help pages you have viewed. You can go home (house icon), print or open help in new window. You
can search help, use the search window. Help can also be browsed through main menu bar at the open
of the page
data()
To load a dataset, for example, the dataset women which gives the average heights and weights for 15
American women aged 30-39.
data(women)
ls()
ls(pattern = "w")
21
Figure 2.4: RStudio Viewing objects in R workspace
The Workspace windows lists the object currently in the R workspace. You can click on each item to view
or edit it. Note women is a data table with dimensions 15 rows x 2 columns, you can click on the table icon
to view it. Z and y are a single value (50). x and y are integer vectors of length 10.
2.5 R Packages
By default, R is packaged with a small number of essential packages, however as we saw there are many
contributed R packages.
1. Some packages are loaded by default with every R session. The libraries included in the Table are
loaded on the R startup.
search()
sessionInfo()
3. To see which packages are installed on your computer, issue the command
library()
22
Within RStudio installed packages can be view in the Package Tab of the lower right panel. You
can tick or select a library to load it in R.
Figure 2.5: RStudio- Viewing the list of all packages that are Installed loaded into a workspace
The list is all installed packages. A tick marked indicates its loaded into the current R session. Clicking on
the package name will open help for that package
23
You can open a CMD shell from RStudio Tools -> Shell or type cmd to run cmd.exe into the Start box
on windows
Figure 2.7: Type the name of the package. Unfortunately RStudio does not display a list of all available
packages
On windows, sometimes I have encountered problems with program installation using RStudio. This is
normally when R tried to write to a folder but doesnt have write permission to this folder. To check this
2. Or check you have write permission to the path where it will write the files (use the following R func-
tion to list your RStudio Paths) rsPref, eval=FALSE= .rs.defaultUserLibraryPath() Path where
RStudio installs packages .rs.rpc.getp ackagei nstallc ontext()Otherusef ulRStudioconf igurationinf o@
24
# Installing and updating R libraries
install.packages("lme4")
update.packages("lme4")
Installation of all packages takes some time and space on your computer. If the name of the package is not
known, you could use task-views help or archives of the mailing list to pinpoint one. Also look on the R
website Task views description of packages (see Additional Notes in Installation which I have provided).
To list all of the packages installed on your computer.
/subsectionCreating an Install R packages script Its useful to create an install package script, to automate
package installation when you wish to update R to a new version.
The easiest way to do this is to take the list of currently install package and send to them to a script, similar
to that at the start of this manual in section
Note if you have Bioconductor packages, the install is different. See the online manual on installing Bio-
conductor packages, and then use biocLite rather than install.packages to install Bioconductor packages.
library(help = lme4)
Many packages, particular those in Bioconductor have vignette documents which provide a case studies or
tutorial on the functions in a package
To see if a package has a vignette, either browse the package help in the Rstudio help window, or online
using help.search
25
2.6.5 Load package into workspace
Once you have installed a package, you do NOT need to re-install it. To load the library in your current R
session use the commands
sessionInfo()
search()
search()
detach(package:lme4)
search()
NOTE: Packages are often inter-dependent, and loading one may cause others to be automatically loaded.
To customize your R environment, you can change options through RStudio File -> Tools -> Options
2.7.2 Rprofile
In addition you can specify preferences for either a site or local installation in Rprofile.site. On Windows,
this file is in the C:\Program Files\R\R-x.y.z\etc directory where x.y.z is the version of R. Or
you can also place a .Rprofile file in the directory that you are going to run R from or in the user home
directory.
At startup, R will source the Rprofile.site file. It will then look for a .Rprofile file in the current working
directory. If it doesnt find it, it will look for one in the users home directory.
There are two special functions you can place in these files. .First( ) will be run at the start of the R session
and .Last( ) will be run at the end of the session. These can be used to load a set of libraries that you use
most.
26
Figure 2.8: Options in RStudio
Within the options window you can change the version of R, your default directory and SVN preferences
# General options
options(tab.width = 2)
options(width = 100)
options(digits = 5)
27
Chapter 3
Objects in R
R creates and manipulates objects: variables, matrices, strings, functions, etc. objects are stored by name
during an R session.
During a R session, you may create many objects, if you wish to list the objects you have created in the
current session use the command
objects()
ls()
rm(x, y, z, junk)
Objects can be thought of as a container which holds data or a function. The most basic form of data is
a single element, such as a single numeric or a character string. However one cant do statistics on single
numbers! Therefore there are many other objects in R.
28
3.2.1 vector
A vector is an ordered collection of numerical, character, complex or logical objects. Vectors are collec-
tion of atomic (same data type) components or modes. For example
# Numeric
vec1 <- 1:10
vec1
## [1] 1 2 3 4 5 6 7 8 9 10
# Character
vec2 <- LETTERS[1:10]
vec2
## [1] "A" "B" "C" "D" "E" "F" "G" "H" "I" "J"
# logical
vec3 <- vec2 == "D"
vec3
## [1] FALSE FALSE FALSE TRUE FALSE FALSE FALSE FALSE FALSE FALSE
In each case above, these vectors have 10 elements, and are of length=10.
3.2.2 Matrix
A matrix is a multidimensional collection of data entries of the same type. Matrices have two dimensions.
It has rownames and colnames.
## [,1] [,2]
## [1,] 1 6
## [2,] 2 7
## [3,] 3 8
## [4,] 4 9
## [5,] 5 10
dim(mat1)
## [1] 5 2
29
colnames(mat1) <- c("A", "B")
rownames(mat1) <- paste("N", 1:5, sep = "")
print(mat1)
## A B
## N1 1 6
## N2 2 7
## N3 3 8
## N4 4 9
## N5 5 10
3.2.3 list
A list is an ordered collection of objects that can be of different modes (e.g. numeric vector, array, etc.).
a <- 20
newList1 <- list(a, vec1, mat1)
print(newList1)
## [[1]]
## [1] 20
##
## [[2]]
## [1] 1 2 3 4 5 6 7 8 9 10
##
## [[3]]
## A B
## N1 1 6
## N2 2 7
## N3 3 8
## N4 4 9
## N5 5 10
##
## $a
## [1] 20
##
## $myVec
## [1] 1 2 3 4 5 6 7 8 9 10
##
## $mat
## A B
30
## N1 1 6
## N2 2 7
## N3 3 8
## N4 4 9
## N5 5 10
##
3.2.4 data.frame
Though a data.frame is a restricted list with class data.frame, it maybe regarding as a matrix with
columns that can be of different modes. It is displayed in matrix form, rows by columns. (Its like an excel
spreadsheet)
## A B
## N1 1 6
## N2 2 7
## N3 3 8
## N4 4 9
## N5 5 10
3.2.5 factor
## [1] "A" "B" "C" "A" "B" "C" "A" "B" "C" "A" "B" "C" "A" "B" "C" "A"
## [17] "B" "C" "A" "B" "C" "A" "B" "C" "A" "B" "C" "A" "B" "C"
## charVec
## A B C
## 10 10 10
## [1] A B C A B C A B C A B C A B C A B C A B C A B C A B C A B C
## Levels: A B C
31
attributes(fac1)
## $levels
## [1] "A" "B" "C"
##
## $class
## [1] "factor"
##
str(fac1)
levels(fac1)
The order of the levels of these categories is not important, but sometimes the order of the factor is important
By default the levels are ordered according to the alphabetical order of the categories.
But this may not be appropriate where factor order in important. For example if we have a survey response
scale of good to poor. Or we are looking at a gradient such as tumor grade 1 to 3. In these cases, the order
is important.
For example, we have a variable of socio-economic status from low to high. By default alphabetical ordering,
high is the lowest and low is the highest level. In order to fix the ordering we need to use levels argument
to indicate the correct ordering of the categories.
levels(fac1)
## [1] low low low medium medium medium high low high
## [10] medium
## Levels: low medium high
32
fac3 <- factor(facA, levels = c("low", "medium", "high"),
ordered = TRUE)
fac3
## [1] low low low medium medium medium high low high
## [10] medium
## Levels: low < medium < high
## [1] low low low medium medium medium high low high
## [10] medium
## Levels: low < medium < high
3.2.7 array
An array in R can have one, two or more dimensions. I find it useful to store multiple related data.frame (for
example when I jack-knife or permute data). Note if there are insufficient objects to fill the array, R recycles
(see below)
## , , 1
##
## [,1] [,2] [,3] [,4]
## [1,] 1 3 5 7
## [2,] 2 4 6 8
##
## , , 2
##
## [,1] [,2] [,3] [,4]
## [1,] 9 11 13 15
## [2,] 10 12 14 16
##
## , , 3
##
## [,1] [,2] [,3] [,4]
## [1,] 17 19 21 23
## [2,] 18 20 22 24
##
33
array(1:23, dim = c(2, 4, 3))
## , , 1
##
## [,1] [,2] [,3] [,4]
## [1,] 1 3 5 7
## [2,] 2 4 6 8
##
## , , 2
##
## [,1] [,2] [,3] [,4]
## [1,] 9 11 13 15
## [2,] 10 12 14 16
##
## , , 3
##
## [,1] [,2] [,3] [,4]
## [1,] 17 19 21 23
## [2,] 18 20 22 1
##
## , , X
##
## A B C D
## Patient1 1 3 5 7
## Patient2 2 4 6 8
##
## , , Y
##
## A B C D
## Patient1 9 11 13 15
## Patient2 10 12 14 16
##
## , , Z
##
## A B C D
## Patient1 17 19 21 23
## Patient2 18 20 22 1
##
34
The most basic and fundamental properties of every objects is its modeand length These are intrinsic
attributes of every object. Examples of mode are logical, numeric, character, list, expression,
name/symbol and function.
Of which the most basic of these are:
integer: an integer
# Numeric
x <- 3
mode(x)
## [1] "numeric"
# Charachter
x <- "apple"
mode(x)
## [1] "character"
x <- 3.145
x + 2 # 5.145
## [1] 5.145
# FALSE, logical
x == 2
## [1] FALSE
x <- x == 2
x
## [1] FALSE
mode(x)
## [1] "logical"
35
# vectors of different modes Numeric
x <- 1:10
mode(x)
## [1] "numeric"
## [1] "numeric"
# Character
x <- LETTERS[1:5]
mode(x)
## [1] "character"
# logical
x <- x == "D"
mode(x)
## [1] "logical"
Quick Exercise
Repeat above, and find the length and class of x in each case.
3.3.1 Dimension
attributes(x)
## $dim
## [1] 2 5
##
In summary
36
Object Modes Allow >1 Modes*
vector numeric, character, complex or logical No
matrix numeric, character, complex or logical No
list numeric, character, complex, logical, function, expression, ... Yes
data frame numeric, character, complex or logical Yes
factor numeric or character No
array numeric, character, complex or logical No
*Whether object allows elements of different modes. For example all elements in a vector or array have to be of the
same mode. Whereas a list can contain any type of object including a list.
# Vector
x.vec <- seq(1, 7, by = 2)
# The function seq is very useful, have a look at the help on seq
# (hint ?seq)
Accessing elements
NOTE Use square brackets to access elements. The number of elements within the square
bracket must equal the dimension of the object.
1. vector [1]
2. matrix [1,1]
3. array with 3 dimensions [1,1,1]
## a
## 1
37
# or if you know the name
x.vec["a"]
## a
## 1
## [1] 5
##
## a -1.2081 5
## b -1.2717 5
## c -0.7870 5
## d 0.3818 5
# or
xMat[, -c(1)]
##
## a -1.2081 5
## b -1.2717 5
## c -0.7870 5
## d 0.3818 5
xMat[xMat[, 1] > 3, ]
## x.vec
## c 5 -0.7870 5
## d 7 0.3818 5
If the object has class data.frame or list, you can use the dollar symbol $ to access elements.
The $ can only access columns of data.frame
38
# Get the vector of 'ind' from 'x.df'
colnames(x.df)
x.df$ind
## [1] 1 3 5 7
x.df[, 1]
## [1] 1 3 5 7
names(newList1)
newList1$a
## [1] 20
# Change the element of 'xMat' in the third row and first column to
# '6'
xMat[3, 1] <- 6
# Replace the second column of 'z.mat' by 0's
z.mat[, 2] <- 0
# Simplest 'sort'
z.vec <- c(5, 3, 8, 2, 3.2)
sort(z.vec)
39
order(z.vec)
## [1] 4 2 5 1 3
Sorting the rows of a matrix. We will use an example dataset in R called ChickWeight. First
have a look at the ChickWeight documentation (help)
Lets take a subset of the matrix, say the first 36 rows.
# ?ChickWeight
ChickWeight[1:2, ]
## by just weight
chickOrd <- chick.short[order(chick.short$weight), ]
chickOrd[1:5, ]
40
x <- matrix(c(2, 1, 1, 3, 0.5, 0.3, 0.5, 0.2), ncol = 2)
# Sort the second column in decreasing order
x1 <- x[order(x[, 2], decreasing = TRUE), ]
# Sort the first column in the partially sorted matrix
x2 <- x1[order(x1[, 1]), ]
Or if both columns are numeric, you negatives sort in the reverse order of positives
## [,1] [,2]
## [1,] 1 0.5
## [2,] 1 0.3
## [3,] 2 0.5
## [4,] 3 0.2
If the values arent known to be numeric, convert them to numeric before sorting
## [,1] [,2]
## [1,] 1 0.5
## [2,] 1 0.3
## [3,] 2 0.5
## [4,] 3 0.2
Note with both of these, missing values NA will be appended to the end of the list
## [1] 4 5 1 3 2
z.vec[order(z.vec)]
41
3.5.3 Creating empty vectors and matrices
To create a empty vector, matrix or data.frame
x1 <- numeric()
x2 <- numeric(5)
x1.mat <- matrix(0, nrow = 10, ncol = 3)
a <- 1:3
length(a) <- 4
a
## [1] 1 2 3 NA
## [1] 1 2 3 NA
print(z)
## [1] 1 2 3 NA
x <- z[!is.na(z)]
print(x)
## [1] 1 2 3
Check to see if a vector has all, any or a certain number of missing values. These create logical
vectors which can be used to filter a matrix or data.frame
all(is.na(z))
## [1] FALSE
42
any(is.na(z))
## [1] TRUE
sum(is.na(z))
## [1] 1
sum(is.na(z)) > 1
## [1] FALSE
43
3.6 Quick recap
R Environment, interface, R help and R-project.org and Bioconductor.org website
installing R and R packages.
assignment <-, =, ->
operators ==, !=, <, >, Boolean operators &, |
Management of R session, starting session, getwd(), setwd(), dir()
Listing and deleting objects in memory, ls(), rm()
R Objects
Object Modes Allow >1 Modes*
vector numeric, character, complex or logical No
matrix numeric, character, complex or logical No
list numeric, character, complex, logical, function, expression, ... Yes
data frame numeric, character, complex or logical Yes
factor numeric or character No
array numeric, character, complex or logical No
*Whether object allows elements of different modes. For example all elements in a vector or array have to be
of the same mode. Whereas a list can contain any type of object including a list.
There are other objects type include ts (time series) data time etc. See the R manual for more
information. All R Objects have the attributes mode and length.
Creating objects; c(), matrix(), data.frame(), seq(), rep(), etc
Adding rows/columns to a matrix using rbind() or cbind()
Sub-setting/Accessing elements in a vector(), matrix(), data.frame(), list() by element name
or index.
44
3.7 Exercise 1
For this exercise we will work on data from a study which examined the weight, height and age of
women. Data from the women Study is available as an R dataset and information about this study
can be found by using R help (hint ?women) which we will read directly from the website URL
http://bcb.dfci.harvard.edu/aedin/courses/R/Women.txt into the object women
Basic tools for reading and writing data are respectively: read.table and write.table. We will go
into further detail about each later today, but first lets read in this file by typing these commands:
45
Chapter 4
So far, we have only analyzed data that were already stored in R. Basic tools for reading and
writing data are respectively: read.table and write.table. We will go into further detail about each.
First we will talk about reading in simple text documents; comma and tab-delimited format text
files, then Excel and import/export from other statistical software.
Figure 4.1: which provides an easy approach to read a text from a local directory or directly from a web
URL
Enter a file location (either local or on the web), and RStudio will make a best guess at
the file format. There are a limited number of options (heading yes or no), separators (comma,
space or tab) etc but these should cover the most common data exchange formats (The R interfaces
RCommander and RExcel also provide rich support for data import of many different file formats
into R)
46
Figure 4.2: The top panel shows the plain text of the file, and the lower panels displays how R is interpreting
the data. Black rows are the column headings
Women<-read.table("Women.txt")
In order to read files that are tab or comma delimited, the defaults must be changed. We also
need to specify that the table has a header row
# Tab Delimited
Women <- read.table("Women.txt", sep = "\t", header = TRUE)
Women[1:2, ]
47
summary(Women)
class(Women$age)
## [1] "integer"
Note by default, character vector (strings) are read in as factors. To turn this off, use the
parameter as.is=TRUE
2. Important options:
header==TRUE should be set to TRUE, if your file contains the column
names
as.is==TRUE otherwise the character columns will be read as factors
sep= field separator character (often comma , or tab eg:
sep=,)
na.strings a vector of strings which are to be interpreted as NA val-
ues.
row.names The column which contains the row names
comment.char by default, this is the pound # symbol, use to turn off
interpretation of commented text.
Note the defaults for read.table(), read.csv(), read.delim() are different. For example, in
read.table() function, we specify header=TRUE, as the first line is a line of headings among
other parameters.
3. read.csv() is a derivative of read.table() which calls read.table() function with the following
options so it reads a comma separated file:
48
# Comma Delimited
Women2 <- read.csv("Women.csv")
Women2[1:2, ]
4. Reading directly from Website You can read a file directly from the web
Files compressed via the algorithm used by gzip can be used as connections created by the
function gzfile, whereas files compressed by bzip2 can be used via bzfile. Suppose your data
is in a compressed gzip or tar.gz file, you can use the R gzfile function to decompress on the
fly. Do this:
49
4.3 Exercise 2
The ToothGrowth data are from a study which examined the growth of teeth in guinea pigs (n=10)
in response to three dose levels of Vitamin C (0.5, 1, and 2 mg), which was administered using two
delivery methods (orange juice or ascorbic acid). Data from the Tooth Growth Study is available
as an R dataset and information about this study can be found by using R help (hint ?ToothGrowth)
1. Download the data set ToothGrowth.xls which is available on the course website. Save it
in your local directory. Open this file ToothGrowth.xls in Excel.
2. Export the data as both a comma or tab delimited text files. In Excel select File -> Save as
and
Tab: select the format Text (Tab delimited) (*.txt).
CSV: select the format CSV (Comma delimited) (*.csv).
50
4.3.1 Importing text files Using scan()
NOTE: read.table() is not the right tool for reading large matrices, especially those with many
columns. It is designed to read data frames which may have columns of very different classes.
Use scan() instead.
scan() is an older version of data reading facility. Not as flexible, and not as user-friendly as
read.table(), but useful for Monte Carlo simulations for instance. scan() reads data into a vector or
a list from a file.
Note by default scan() expects numeric data, if the data contains text, either specify what=text
or give an example what=some text.
Other useful parameters in scan() are nmax (number of lines to be read) or n (number of items
to be read).
51
4.4 Writing Data table from R
1. Function sink() diverts the output from the console to an external file
2. Writing a data matrix or data.frame using the write.table() function write.table() has similar
arguments to read.table()
3. Important options
4. Output to a webpage
The package R2HTML will output R objects to a webpage
52
myPath <- file.path("C://Project1")
file.exists(myPath)
## [1] FALSE
It is better to expand a path using file.path() rather than paste() as file.path() will expand the
path with delimiting characters appropriate to the operating system in use (eg / unix, \, windows
etc)
Use file.exists() to test if a file can be found. This is very useful. For example, use this to test
if a file exists, and if TRUE read the file or you could ask the R to warn or stop a script if the file
does not exist
if (!file.exists(myfile)) {
print(paste(myfile, "cannot be found"))
} else {
Women <- read.table(myfile, sep = "\t", header = TRUE)
Women[1:2, ]
}
53
4.5 Viewing Data
After you use read.table or other function to import data into you, you will need to quickly review
the data to ensure it important correctly
# Data
head(women)
tail(women)
# View the head and tail (15 line) of the function cor which
# calculates correlation
head(cor)
##
## 1 function (x, y = NULL, use = "everything", method = c("pearson",
## 2 "kendall", "spearman"))
## 3 {
## 4 na.method <- pmatch(use, c("all.obs", "complete.obs", "pairwise.com
## 5 "everything", "na.or.complete"))
## 6 if (is.na(na.method))
tail(cor, n = 15)
54
##
## 99 x2 <- rank(x2[ok])
## 100 y2 <- rank(y2[ok])
## 101 r[i, j] <- if (any(ok))
## 102 .Internal(cor(x2, y2, 1L, method == "kendall"))
## 103 else NA
## 104 }
## 105 }
## 106 rownames(r) <- colnames(x)
## 107 colnames(r) <- colnames(y)
## 108 if (matrix_result)
## 109 r
## 110 else drop(r)
## 111 }
## 112 }
## 113 }
4.5.2 str
The function str is very useful, it displays the structure of an R object
str(women)
str(1:10)
## int [1:10] 1 2 3 4 5 6 7 8 9 10
str(cor)
55
4.5.3 The function Summary
The function summary can be applied to a vector, factor,matrix,data.frame, list etc. and will be
summarized the object. The function will generate a summary specialized for the class of the
object.
methods(summary)
For a vector, data.frame or matrix its provides information data distribution (min, quantile,
mean, max etc)
summary(1:10)
summary(women)
56
vec <- sample(c("old", "young"), 20, replace = TRUE)
summary(vec)
summary(factor(vec))
## old young
## 10 10
4.5.4 Table
The function table will tabulate the elements of a character vector
table(vec)
## vec
## old young
## 10 10
class(vec)
## [1] "character"
I have included a lot more on cross-tabulation and generating summaries on data in chapter 9
57
4.6 Exercise 3
1. Use read.table() to read the space separated text file WomenStats.txt directly from the website
http://bcb.dfci.harvard.edu/aedin/courses/R/WomenStats.txt, Call
this data.frame women.
2. Change the rownames to be the letters of the alphabet eg A, B C D etc
3. Write out this file as a tab delimited file using write.table()
4. Read this into R using read.table(). What parameters need modifying to read the data as a
tab-delimited file?
58
4.7 Importaing Data from other Software
4.7.1 Reading data from Excel into R
There are several packages and functions for reading Excel data into R, however I normally export
data as a .csv file and use read.table().
However if you wish to directly load Excel data, here are many the options available to you.
See the section on Importing-from-other-statistical-systems in the webpage http://cran.
r-project.org/doc/manuals/R-data.html for more information
1. xslx seems to be the simplest option at the moment
library(xlsx)
ww <- read.xlsx(file = "Women.xlsx", sheetIndex = 1)
read.xlsx accepts .xls and .xlsx format. You must include a worksheet name or number. It is
optional to specify a row or column index to indicate a section of a Worksheet
2. There is also a packages call XLConnect which is similar
require(XLConnect)
wb <- loadWorkbook("Women.xlsx", create = TRUE)
WW <- readWorksheet(wb, sheet = 1)
Or you can read direct from a connection, calling the file directly.
3. RExcel R can be ran from within Excel on Windows using RExcel (http://rcom.univie.
ac.at/). This add a menu to Excel that allows you to call R functions from within Excel.
RExcel is part of the much large Statconn project.
4. RODBC library. We are not sure it will not work with .xlsx files. See the vignette for more
information
library(RODBC)
RShowDoc("RODBC", package = "RODBC")
The following RODBC function works under windows, but may have issues under MacOS or
Linux as may need to install ODBC drivers.
channel<-odbcConnectExcel("ToothGrowth.xls")
#list the spreadsheets
sqlTables(channel)
59
5. The gdata library function read.xls()
Perl must be installed on your computer in order for these to work, as it uses the Perl functions
xls2csv or xls2tab.
In R
library(Hmisc)
mydata <- sasxport.get("c:/mydata.xpt")
SAScii. Anthony Joseph Damico recently announced SAScii is a new packages to parse SAS
input code to read.fwf However although they stated the code below should work, I have been
not so successful with it in my hands.
require("SAScii")
60
#store the NHIS file as an R data frame!
NHIS10_personsx_df <- read.SAScii(NHIS10_personsx_SASInst, NHIS10_person
Recently Xin Wei of Roche Pharmaceuticals published a SAS macro called PROC_R that
may potentially ease integrating R and SAS (reference Xin Wei PROC_R: A SAS Macro that
Enables Native R Programming in the Base SAS Environment J. Stat Software. Vol. 46,
Code Snippet 2, Jan 2012) which allows you to put R code within a SAS macro.
%include "C:\aedin\sasmacros\Proc_R.sas";
%Proc_R (SAS2R =, R2SAS =);
Cards4;
******************************
***Please Enter R Code Here***
******************************
;;;;
%Quit;
4.8.2 S
PSS
From SPSS, save SPSS dataset in transport format
get file='c:\mydata.sav' .
export outfile='c:\mydata.por' .
In R
library(Hmisc)
mydata <- spss.get("c:/mydata.por", use.value.labels = TRUE)
library(foreign)
mydata <- read.dta("c:/mydata.dta")
61
4.9 From a Database Connection
There is also support for database connectivity including for mySQL, Oracle and specialized file
formats including network Common Data Form (netCDF) etc. See http://cran.r-project.
org/doc/manuals/R-data.html for more details.
Note installation of RMySQL or ROracle is simple on Linux or Mac, but maybe complex on
MSWindows, as there is no binary file. See the ReadMe associated with the package on the R
website
62
4.10 Sampling and Creating simulated data
1. seq and rep. we have already seen the function seq and rep which generate a sequence or
repeat elements.
2. Create data from a specific distribution
Often we want to sample data from a specific distribution, also sometimes called simulating
data. This data is usually used to test some algorithm or function that someone has written.
Since the data is simulated, you know where it came from and so what the answer should be
from your algorithm or function. Simulated data lets you double-check your work.
Each distribution has 4 functions associated with it:
For example, rnorm( ), dnorm( ), pnorm( ), and qnorm( ) give random normals, the normal
density (sometimes called the differential distribution function), the normal cumulative distri-
bution function (CDF), and the inverse of the normal CDF (also called the quantile function),
respectively.
Almost all of the other distributions have similar sets of four functions. The r versions are
rbeta, rbinom, rcauchy, rchisq, rexp, rf, rgamma, rgeom, rhyper, rlogis, rlnorm, rmultinom,
rnbinom, rnorm, rpois, rsignrank, rt, runif, rweibull, and rwilcox (there is no rtukey because
generally only ptukey and qtukey are needed).
For example, generate 5 observations from a normal distribution with mean 0 and stdev 1, or
10 observation with a mean of 20 and a stdev of 2
rnorm(5, 0, 1)
rnorm(10, 6, 2)
## [1] 6.488 8.059 7.217 9.168 4.904 8.448 5.516 4.433 4.003 4.582
For most of the classical distributions, these simple function provide probability distribution
functions (p), density functions (d), quantile functions (q), and random number generation
(r). Beyond this basic functionality, many CRAN packages provide additional useful distri-
butions. In particular, multivariate distributions as well as copulas are available in contributed
packages. See http://cran.r-project.org/web/views/Distributions.html
and http://cran.r-project.org/doc/manuals/R-intro.html#Probability-distri
for more information.
3. Sample from existing data
The second type of simulation you may wish to perform is to bootstrap or permute existing
data. In bootstrapping one generally follows the same basic steps
(a) Resample a given data set a specified number of times
(b) Calculate a specific statistic from each sample
(c) Find the standard deviation of the distribution of that statistic
63
The function sample() will resample a given data set with or without replacement
sample(1:10)
## [1] 6 1 4 3 10 2 8 9 7 5
## [1] 10 4 8 10 4 1 5 3 8 5
You can also add weights to bias selection or probability of selecting of a certain subset. For
example bootstrap sample from the same sequence (1:10) with probabilities that favor the
numbers 1-5
## [1] 0.25 0.25 0.25 0.25 0.25 0.05 0.05 0.05 0.05 0.05
## [1] 9 5 2 2 1 4 8 5 3 3
4.11 Exercise 4
1. Create the vector which contains the first 20 letters of the alphabet and the sequence of
number 0:200 in increments of 10 (hint use seq()).
2. Use sample() to randomize the order of the vector.
3. Use the function cat() to write this vector to a file called myVec.txt.
4. Use scan() to read the first 10 items in the file, what value do you give to the parameter
what. Compare running scan() with different data types; eg: what=text, what=123 and
what=TRUE
64
Chapter 5
## [1] 20.6
myMean(testVec)
## [1] 20.6
65
5. A more complex example
Example of a function twosam: takes as arguments two vectors y1 and y2, calculates
the 2-sample t-test statistic (assuming equal variance), and returns the t-statistic
x <- 9
if (x > 0) sqrt(x) else sqrt(-x)
## [1] 3
## [1] 3
66
spread <- function(x, type) {
switch(type, sd = sd(x), mad = mad(x), IQR = IQR(x)/1.349)
}
samp <- rnorm(50)
spread(samp, 2)
## [1] 0.8806
spread(samp, "IQR")
## [1] 1.108
Why IQR(x)/1.349 ? In a normal distribution 50% of the data (between 0.25 and 0.75 quartiles). So the distance between the
two quartiles IQR(x) = quantile(x,0.75) quantile(x,0.25). For a normal distribution IQR is qnorm(0.75) - qnorm(0.25) 1.349.
Therefore IQR/1.349 is an estimator of the standard deviation of a normal distribution.
where i is the loop variable, expr1 is usually a sequence of numbers, and expr2 is an
expression.
## [1] 1
## [1] 4
## [1] 9
## [1] 16
## [1] 25
x <- 1
y <- 16
while (x2 < y) {
cat(x, "squared is ", x2, "\n") # print x and sq(x)
x <- x + 1
}
67
## 1 squared is 1
## 2 squared is 4
## 3 squared is 9
A word of caution, it is easy to write a while() loop that doesnt terminate, in which case your
script may go into a never-ending cycle. Therefore if possible, write a for() loop in preference to a
while() loop.
Iterative for and while loops in R are sometimes memory intensive, and functions such as apply,
sweep or aggregate should be should instead. For a comparision of the computational efficiency of
For versus apply loops see section 5.8.2
1. apply
appply() applies a function over the rows or columns of a matrix. The syntax is
where X: array, matrix or data.frame; MARGIN: 1 for rows, 2 for columns, c(1,2) for both;
FUN: one or more functions; ARGs: possible arguments for function
For example, lets go back to the example dataset women which we loaded from the web.
summary(women)
colMeans(women)
apply(women, 2, mean)
68
testEq <- all(rowMeans(women) == apply(women, 1, mean))
print(testEq)
## [1] TRUE
2. tapply
tapply() is a member of the very important apply() functions. It is applied to ragged arrays,
that is array categories of variable lengths. Grouping is defined by vector.
Example:
## over35 under35
## 5 10
69
## $over35
## Min. 1st Qu. Median Mean 3rd Qu. Max.
## 120 135 139 142 150 164
##
## $under35
## Min. 1st Qu. Median Mean 3rd Qu. Max.
## 115 124 130 134 145 159
##
## Case A1 B1 C1
## 1 f -0.05922 1 1
## 2 d -0.50757 2 2
## 3 g 0.92258 3 3
## 4 i -0.05271 4 4
## 5 x -2.60035 5 5
## 6 a 1.16796 6 1
## 7 v -2.52888 7 2
70
## 8 q -0.22759 8 3
## 9 m -0.86923 9 4
## 10 o 0.05612 10 5
## A1 Case D1 E1 B1
## 1 1 k 3.263 1 Non-Smoker
## 2 3 v 3.864 2 Non-Smoker
## 3 5 d 4.993 3 Non-Smoker
## 4 7 z 5.525 4 Non-Smoker
## 5 9 n 2.906 5 Smoker
## 6 1 w 2.989 1 Smoker
## 7 3 r 4.262 2 Smoker
## 8 5 u 4.504 3 Smoker
## 9 7 m 5.215 4 <NA>
## 10 9 p 4.851 5 <NA>
5.5 Exercise 5
1. Write a for loop printing the consecutive powers of 2, from 0 to 10
2. Write a while loop printing the consecutive powers of 2, less than 1000
71
5.6 Functions for parsing text
There are many functions with R for parsing text. We will cover some of this here.
To search for text within an R vector, use grep. It uses the same regular expression patterns
as perl is you set perl=TRUE
grep("A", LETTERS)
## [1] 1
## [1] "A"
To search for a string begining with Ma or ending in the letters d or v use regular expression
pattern seach. Here means the start and
## [1] "Mary"
## [1] "Had"
R has full support regular expresion (see ?regex) synthax. For example to search for words
containing the letter a. Here refers to any alphanumeric character and + means one or more
whereas means zero or more.
72
sub("B", "A", LETTERS)
## [1] "A" "A" "C" "D" "E" "F" "G" "H" "I" "J" "K" "L" "M" "N" "O" "P"
## [17] "Q" "R" "S" "T" "U" "V" "W" "X" "Y" "Z"
a <- date()
strsplit(a, " ")
## [[1]]
## [1] "Fri" "Dec" "14" "04:03:23" "2012"
##
strsplit(a, "J")
## [[1]]
## [1] "Fri Dec 14 04:03:23 2012"
##
## [1] "list"
b <- unlist(b)
class(b)
## [1] "character"
For special characters you need to precede them with a double back slash
73
a <- "aedin@jimmy.harvard.edu"
strsplit(a, "\\.")
## [[1]]
## [1] "aedin@jimmy" "harvard" "edu"
##
## done
IQR(xx)
## [1] 11.5
args(IQR)
body(IQR)
IQR
74
Sometimes, functions dont appear to be visible. In this case, use methods or getAnywhere to
view the function.
Someone a function may have different version, and may depend on the class of the input
object. In this case functions will be called functionname.class. In the case below, mean will call
a the function mean.Data or mean.difftime if the input object is a Date, difftime respectivel
mean
methods(mean)
Now we can see the code for mean, as we know the full name of the function is mean.default.
mean.default
Some function are Non visible that means you can see the code. When you These are hidden
in the package namespace. use the function methods, non visible functions are marked by an
asterisk.
75
`?`(t.test)
t.test
methods(t.test)
stats:::t.test.default
To reduce the output and save paper in the manual, we will just view the first 5 and last 10 lines
of the function.
head((stats:::t.test.default), 5)
##
## 1 function (x, y = NULL, alternative = c("two.sided", "less", "greater"),
## 2 mu = 0, paired = FALSE, var.equal = FALSE, conf.level = 0.95,
## 3 ...)
## 4 {
## 5 alternative <- match.arg(alternative)
print("truncated...")
## [1] "truncated..."
tail((stats:::t.test.default), 10)
##
## 102 names(mu) <- if (paired || !is.null(y))
## 103 "difference in means"
## 104 else "mean"
## 105 attr(cint, "conf.level") <- conf.level
## 106 rval <- list(statistic = tstat, parameter = df, p.value = pval,
## 107 conf.int = cint, estimate = estimate, null.value = mu,
## 108 alternative = alternative, method = method, data.name = dname)
## 109 class(rval) <- "htest"
## 110 return(rval)
## 111 }
There are some functions that you will not be able to see using these commands. These are most
likely written in object orientated R (called S4). Much of Bioconductors functions are written in
S4. However a full discussion of S4 functions is beyond the scope of this course.
76
5.8.1 Making scripts work across your computers or platforms
I work on several computers, each have fairly different directory structures. But by setting a con-
sistent directory structure within the home directory I can sync code between computers and have
them run properly on each one since where I run my R projects have similar directory structures.
The R command file.path will automatically add in the correct slashes for windows, mac or
Linux
## [1] "home/projects/Dec01"
The command path.expand will expand a path name. You can combine it with the tilde symbol
( ) which is the shortcut to your home drive (on any operating system)
path.expand("")
myhome <- path.expand("")
newdir <- file.path(path.expand(""), "Rwork", " colonJan13")
setwd(newdir)
Sys.getenv("HOME")
## [1] "C:/Users/aedin/Documents"
Here pathName is the directory you wish to set as your new home directory
Sys.setenv(HOME = "pathName")
Note apply is much more computational efficient that a for loop. But if you can use built in
functions like rowMeans or colMeans these are quicker still
77
myMA <- matrix(rnorm(1000000), 100000, 10, dimnames=
list(1:100000, paste("C", 1:10, sep="")))
78
5.8.3 Handling missing values
mean(a)
## [1] NA
## [1] 5.5
79
5.9 Exercise 6: Parsing Real Data - World Population Data from Wikipedia
We demonstrated how to get data tables a URL. This means we can retrieve data from most any
table on the web. The function readHTMLTable is very flexible for this. Please retrieve the table
entitled Estimated world population at various dates (in millions) (Table 13) from
http://en.wikipedia.org/wiki/World_population.
require(XML)
## [1] "NULL"
## [2] "toc"
## [3] "NULL"
## [4] "World population milestones (USCB estimates)"
## [5] "The 10 countries with the largest total population:"
## [6] "10 most densely populated countries (with population above 1 million)"
## [7] "Countries ranking highly in terms of both total population (more than 15 million p
## [8] "NULL"
## [9] "UN (medium variant 2010 revision) and US Census Bureau (December 2010) estimates[
## [10] "UN 2008 estimates and medium variant projections (in millions)[97]"
## [11] "World historical and predicted populations (in millions)[101][102]"
## [12] "World historical and predicted populations by percentage distribution[101][102]"
## [13] "Estimated world and regional populations at various dates (in millions)"
## [14] "Starting at 500 million"
## [15] "Starting at 375 million"
## [16] "NULL"
## [17] "NULL"
## [18] "NULL"
## [19] "NULL"
1. First tidy the data. The data are factors, its easier to edit data that are character. Apply
as.character to each column
2. Remove rows with dates before 1750. Remove the additional header in row 32.
3. Remove column 9 notes
4. Use the sub to remove the comma in the data values
5. convert the data to numeric
6. In what year did the population of Europe, Africa and Asia exceed 500 million?
7. Bonus: Plot the population growth of the World, Africa or Europe since 1750. Given this
plot, would you guess that the the population of the World, Africa or Europe would be more
likely to double again before the end of 21st century?
80
5.9.1 Writing functions: More on arguments
We are equipped now with all basic tools we need for writing functions. We include a few tips on
arguments to functions.
1. Function arguments: Default values In many cases arguments have default values. For exam-
ple the qnorm(x, mean = 0, SD = 1,lower.tail = TRUE, log.p = FALSE ) has default values
for the mean, standard deviation, cdf calculation and probabilities on the original scale.
81
Formal parameters (appear in the argument list of the function)
Local variables (values are determined by the evaluation of expressions in the body of
the functions)
Free variables (neither of the above)
In this example: x - formal parameter, y - local variable, z - free variable.
Example:
fn <- function(x) {
y <- 2 * x
print(x)
print(y)
print(z)
}
z <- 2
x <- 4
fn(x = 2)
## [1] 2
## [1] 4
## [1] 2
2. Lexical scope.
Example: function called cube.
## [1] 8
n <- 4
cube(2)
## [1] 8
82
5.11 Options for Running memory or CPU intensive jobs in R
5.11.1 Distributed computing in R
There are two ways to split or distribute a big job. The simplest it to send jobs to different pro-
cessors on the same machine (assuming it has multiple cores, which most new machines do). The
second option is to split or parallelize a job across many machines or a cluster of machines. For
both of these, see the Bioconductor package parallel which builds upon the older R packages snow
and multicore
To install parallel use the Bioconductor package installer, BiocInstaller
library(BiocInstaller)
biocLite("parallel")
The package parallel has many functions which work like apply to distribute a computation.
For example use mclapply just like lapply to split a job over 4 cores.
library(parallel)
system.time(mclapply(1:4, function(i) mc.cores <- 4))
mclapply is a parallelized version of lapply, and will not work on windows (as far as I know) but
on Windows you can use functions parLapply, clusterApply and clusterApplyLB all in the parallel
package.
The packages has several functions for different types of apply loops including parLapply,
parSapply, and parApply which are parallel versions of lapply, sapply and apply respectively.
library(parallel)
cl <- makeCluster(3)
parLapply(cl, 1:3, sqrt)
stopCluster(cl)
There are several other packages for distributed computing see the reviews of R packages on
CRAN task views http://cran.r-project.org/web/views/HighPerformanceComputing.
html. I have received recommendations on R packages biglm, ff and bigmemory.
83
If you wish to set up your own instance, Louis Alsett at Trinity College Dublin provides
an RStudio Server Amazon Machine Image (AMI) which can install to your Cloud account see
http://www.louisaslett.com/RStudio_AMI/.
For more information about distributed computing on the Cloud based including using Hadoop
(which I think is used by Revolution Analytics) see the recent book parallel R by Q Ethan McCal-
lum http://www.amazon.com/Parallel-R-Q-Ethan-McCallum/dp/1449309925
84
5.12 Efficient R coding
5.12.1 What is an R script
A R script is simply a text file, with R commands. There are two ways to call these R commands,
start R and using the R function source, or at the command line using R CMD BATCH
#####################
### Author: Mr Bob Parr
### Date: 2011-01-20
### Version: 1.0
### License: GPL (>= 3)
###
### Description: survival analysis function
#####################
You can save this script in a file named censortime.R in your working directory. If you
want to define this function in your workspace, just type source("censortime.R").
Of course, an R script may contain more than functions, it may also contain any analytical
pipeline. Here is another example:
85
#####################
### Author: Mr Bob Parr
### Date: 2011-01-20
### Version: 1.0
### License: GPL (>= 3)
###
### Description: Script fitting a Cox model on the colon data
### and writing the coefficients in a txt file
#####################
## load library
library(survival)
Save this script in a file named coxColon.R in your working directory. you can run it from
your R session using the command source(coxColon.R) or you can run it in batch mode from a
command line (e.g., shell console) using the command R CMD BATCH coxColon.
1. Indentation
Use 4 spaces for indenting. No tabs.
No lines longer than 80 characters. No linking long lines of code using ;
2. Variable Names
Use camelCaps: initial lowercase, then alternate case between words.
3. Function Names
Use camelCaps: initial lower case, then alternate case between words.
In general avoid ., as in some.func
Whilst beyond the scope of this class, R packages are written to either S3 or S4 stan-
dards. In the S3 class system, some(x) where x is class func will dispatch to this func-
tion. Use a . if the intention is to dispatch using S3 semantics.
4. Use of space
Always use space after a comma. This: a, b, c. Not: a,b,c.
86
No space around = when using named arguments to functions. This: somefunc(a=1,
b=2), not: somefunc(a = 1, b = 2).
Space around all binary operators: a == b.
5. Comments
Use ## to start comments.
Comments should be indented along with the code they comment.
6. Misc
Use "<-" not = for assignment.
7. For Efficient R Programming, see slides and exercises from Martin
http://www.bioconductor.org/help/course-materials/2010/BioC2010/
8. R packages which tidy your code There is a package called formatR https://github.
com/yihui/formatR/wiki/ which will format all R script in a folder, indenting loops,
convert the = to -> etc. See its wiki pages above if you are interesting in testing it.
5.12.6 system.time
If you wish to check the efficient of your code to see how long it is taking to run, use the function
system.time which gives the compute time for a function
system.time(rowMeans(df))
87
## user system elapsed
## 0.01 0.00 0.01
system.time()
Writing R packages
Once you have written all your functions in one or several R files, you can use the function pack-
age.skeleton to generate the necessary directories and empty help pages for your package.
package.skeleton(name = "myFirstRPackage")
Hint: all the packages on CRAN and BioConductor are open source, so you can easily download
the source of any package to take a closer look at it. It may be extremely insightful to see how
experienced R developers implemented their own packages.
88
Using SVN
RStudio v0.96 contains an easy interface to subversion (either GIT or SVN), but here is a detailed
guide to using svn
http://tortoisesvn.net/docs/release/TortoiseSVN_en/tsvn-repository.
html#tsvn-repository-create-tortoisesvn
Step 1. Create local SVN repository
1. Open the windows explorer
2. Create a new folder and name it e.g. SVNRepository
3. Right-click on the newly created folder and select TortoiseSVN Create Repository here....
4. A repository is then created inside the new folder. Dont edit those files yourself!!!. If you
get any errors make sure that the folder is empty and not write protected.
5. For Local Access to the Repository you now just need the path to that folder. Just remember
that Subversion expects all repository paths in the form file:///C:/SVNRepository/. Note the
use of forward slashes throughout.
6. So far this is an empty repository, even though Subversion has created several directories and
files! We need to fill it with our project files and connect it with our working project directory
Step 2: Initial import.
1. Somewhere in your hard drive create a directory (e.g. tmp) with the following three subdi-
rectories:
C:\tmp\new\branches
C:\tmp\new\tags
C:\tmp\new\trunk
2. Backup and Tidy your existing scripts and project files (C:\Projects\MyProject). (ie
delete unnecessary files)
3. Copy the contents of \MyProject into the trunk sub-directory (C:\tmp\new\trunk).
4. Import the new directory into the repository (Right-click/TortoiseSVN/Import). Select
URL as file:///C:/SVNRepository/Myproject (forward slashes!)
5. To see it works, right mouse click start TortoiseSVN/Repo-browser... see your
Imported files.. Happy days. Now you have an SVN with all your files
Step 3. Using SVN
1. Now we have created the SVN, the trick is to use it!!! Start by checking out your data. Create
a new scripts directory (or go back to your old one and delete its contents). And right mouse
click and select SVN Checkout
2. To use the SVN Sending (checking in) your changes to the repository: Right-click on selected
files then SVN Commit
89
3. To add new files to the repository. This is a two step process: , first Right-click on selected
files then TortoiseSVN/Add Then Right-click on selected files then SVN Commit
4. If you wish to delete files (remember the SVN will always have a history of them) use Tor-
toiseSVN/Delete
5. Happy Subversioning!
90
Chapter 6
Introduction to graphics in R
To start lets look at the basic plots that can be produced in R using the demo() function
demo(graphics)
On start up, R initiates a graphics device driver which opens a special graphics window for the
display of interactive graphics. If a new graphics window needs to be opened either win.graph() or
windows() command can be issued.
Once the device driver is running, R plotting commands can be used to produce a variety of
graphical displays and to create entirely new kinds of display.
Plot of Vector(s)
1. One vector x (plots the vector against the index vector)
x <- 1:10
plot(x)
91
10
8
6
x
4
2
2 4 6 8 10
Index
30 20 10 0 10 20 30
x
Plot of data.frame elements If the first argument to plot() is a data.frame, this can be as
simply as plot(x,y) providing 2 columns (variables in the data.frame).
92
Lets look at the data in the data.frame airquality which measured the 6 air quality in New
York, on a daily basis between May to September 1973. In total there are 154 observation
(days).
airquality[1:2, ]
0 100 250 60 70 80 90 0 10 20 30
100
Ozone
50
0
250
Solar.R
100
0
20
15
Wind
10
5
60 70 80 90
Temp
9
8
Month
7
6
5
30
20
Day
10
0
0 50 100 5 10 15 20 5 6 7 8 9
Note most plotting commands always start a new plot, erasing the current plot if necessary.
93
Well discuss how to change the layout of plots so you can put multiple plots on the same
page a bit later. But a simple way to put multiple plots in the same window is by splitting the
display using mfrow
Note if you give plot a vector and factor plot(factor, vector) or plot(vector factor) it will
produce a boxplot.
150
90
100
80
Ozone
Temp
70
50
60
0 50 100 150 5 6 7 8 9
Ozone month
plot(airquality$Ozone~factor(airquality$Month)
detach(airquality)
94
6.2 Exercise 7
Using the ToothGrowth data we read earlier. Please draw the following plot
35
35
30
30
25
25
Tooth Length
Tooth Length
20
20
15
15
10
10
5
5
OJ 0.5
OJ 1
OJ 2
VC 0.5
VC 1
VC 2
OJ VC
Treatment Treatment and Dose
95
6.2.1 Arguments to plot
axes=FALSE Suppresses generation of axes-useful for adding your own custom axes with the
axis() function. The default, axes=TRUE, means include axes.
type= The type argument controls the type of plot produced, as follows:
type=p Plot individual points (the default)
type=l Plot lines
type=b Plot points connected by lines (both)
type=o Plot points overlaid by lines
type=h Plot vertical lines from points to the zero axis (high-density)
type=n No plotting at all. However axes are still drawn (by default) and the coordinate system is
set up according to the data. Ideal for creating plots with subsequent low-level graphics
functions.
xlab=string
ylab=string Axis labels for the x and y axes. Use these arguments to change the default labels,
usually the names of the objects used in the call to the high-level plotting function.
main=string Figure title, placed at the top of the plot in a large font.
sub=string Sub-title, placed just below the x-axis in a smaller font.
xp <- 1:100/100
yp <- 3 * xp2 - 2 * xp + rnorm(100, sd = 0.2)
96
Plot type: l Plot type: b
1.5
1.5
1.0
1.0
0.5
0.5
yp
yp
0.0
0.0
0.5
0.5
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
xp xp
Plot type: o Plot type: h
1.5
1.5
1.0
1.0
0.5
0.5
yp
yp
0.0
0.0
0.5
0.5
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
xp 0.00 0.25
xp
0.50 0.75 1.00
R simple plot
1.5
1.2
1.0
0.6
0.5
values
yp
0.0
0.0
0.5
0.6
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
index xp
boxplot(airquality)
97
300
250
200
150
100
50
0
Note if you give plot a vector and factor plot(factor, vector) or plot(vector factor) it will
produce a boxplot.
98
Equivalent plots
150
150
100
100
ozone
ozone
50
50
0
0
5 6 7 8 9 5 6 7 8 9
month month
boxplot(airquality$Ozone~airquality$Month plot(factor(airquality$Month), airquality$Ozone
150
airquality$Ozone
100
50
0
5 6 7 8 9
factor(airquality$Month)
plot(airquality$Ozone~factor(airquality$Month)
barplot Plot a bar plot of the mean ozone quality by month. First use tapply to calculate the
mean of ozone by month
99
Mean Ozone by month
50
40
30
20
10
0
5 6 7 8 9
## NA
100
## The following object(s) are masked from 'package:utils':
##
## object.size
plotmeans(state.area state.region)
plotmeans(state.area state.region, connect = FALSE)
101
200000
200000
150000
150000
state.area
state.area
100000
100000
50000
50000
Note on Windows, the gplots package will warn if it cant find the programming language
Perl. The above functions will still work, but this notice can be removed by installing Perl
from http://www.perl.org. Linux and Mac OS X users wont see this error as both
already have Perl installed. But it is not installed by default on windows.
To fix this notice, install Perl. I tend to select the active state distribution of Perl. It will
install into C:\perl Then run the following:
installXLSXsupport(perl = "C:/perl/bin/perl.exe")
102
To see more complex examples of plotting confidence intervals or standard error on plots, see
http://wiki.stdout.org/rcookbook/Graphs/Plotting%20means%20and%
20error%20bars%20(ggplot2)/
pie chart
6
7
5
hist(x)- histogram of a numeric vector x with a few important optional arguments: nclass=
for the number of classes, and breaks= for the breakpoints
0.20
30
Frequency
Density
20
0.10
10
0.00
0
8 6 4 2 0 2 4 6 5 0 5
xt N = 100 Bandwidth = 0.4973
103
# generate 2 populations of with different mean and sd
set.seed(54321)
v1 <- rnorm(100, 0, 1)
v2 <- rnorm(100, 10, 7)
5 0 5 10 15 20 25
c(v1, v2)
3D scatterplot
require(scatterplot3d)
data(trees)
trees[1:2,]
104
## Girth Height Volume
## 1 8.3 70 10.3
## 2 8.6 65 10.3
105
Example of scatterplot3d plot: Tree Data
80
70
60
50
Volume
40
Height
90
85
30
80
75
20
70
65
10
60
8 10 12 14 16 18 20 22
Girth
detach(trees)
The R function venn - draws a venn diagram. Input is a list. It will draw a venn diagram
showing the intersect between 2-6 vectors in a list.
require(gplots)
sample(LETTERS, 10)
## [1] "S" "F" "M" "U" "J" "D" "X" "G" "B" "Y"
106
## $Lucy
## [1] "C" "M" "I" "F" "W" "Z" "V" "R" "X" "U"
##
## $Sally
## [1] "Z" "C" "A" "T" "S" "L" "W" "J" "Q" "M"
##
## $Kate
## [1] "F" "D" "C" "M" "P" "N" "H" "E" "Q" "G"
##
venn(tt)
Kate
6
1 1
5 5
Lucy Sally
2
Plot 4 intersections
List 2 List 3
List 1 2
0
1 List 4
2 2
1 2
1 1
0
5 4
0 0
1
107
Color plots
require(venneuler)
plot(venneuler(xx))
List 3 List 4
List 1
List 2
108
tt <- lapply(1:5, function(x) sample(LETTERS, 10))
names(tt) <- sapply(tt, function(x) paste(sort(x), collapse = ""))
venn(tt)
EFGPRSWXYZ
1
DEFGHMNPQX
BCFJKPQWYZ 1 1
1 1 1
0 0
1
1 1
0
1 0
1 1 0
2
0 0
0
3 0 1
0 1
0 0
0
2
3
BCFGIKLNUZ FHJMOSTWXZ
109
6.3 Customize plot with low-level plotting commands
Sometimes the default plot doesnt produce the plot you desire. In this case, low-level plotting
commands can be used to edit or add extra information (such as points, lines or text) to the current
plot. Some of the more useful low-level plotting functions are:
points(x, y)
lines(x, y) Adds points or connected lines to the current plot.
text(x, y, labels, ...) Add text to a plot at points given by x, y. Normally labels is an integer or char-
acter vector in which case labels[i] is plotted at point (x[i], y[i]). The default is 1:length(x).
Note: This function is often used in the sequence
The graphics parameter type=n suppresses the points but sets up the axes, and the text()
function supplies special characters, as specified by the character vector names for the points.
abline(a, b) Adds a line of slope b and intercept a to the current plot.
abline(h=y) Adds a horizontal line
abline(v=x) Adds a vertical line
polygon(x, y, ...) Draws a polygon defined by the ordered vertices in (x, y) and (optionally) shade
it in with hatch lines, or fill it if the graphics device allows the filling of figures.
legend(x, y, legend, ...) Adds a legend to the current plot at the specified position. Plotting char-
acters, line styles, colors etc., are identified with the labels in the character vector legend.
At least one other argument v (a vector the same length as legend) with the corresponding
values of the plotting unit must also be given, as follows:
legend( , fill=v) Colors for filled boxes
legend( , col=v) Colors in which points or lines will be drawn
legend( , lty=v) Line styles
legend( , lwd=v) Line widths
legend( , pch=v) Plotting characters
title(main, sub) Adds a title main to the top of the current plot in a large font and (optionally) a
sub-title sub at the bottom in a smaller font.
axis(side, ...) Adds an axis to the current plot on the side given by the first argument (1 to 4,
counting clockwise from the bottom.) Other arguments control the positioning of the axis
within or beside the plot, and tick positions and labels. Useful for adding custom axes after
calling plot() with the axes=FALSE argument.
To add Greek characters, either specify font type 5 (see below) or use the function expression
110
A random eqn x =
1.0
n
0.5
cos(x)
0.0
0.5
1.0
30 20 10 0 10 20 30
x
This is the subtitle
attach(cars)
plot(cars, type = "n", xlab = "Speed [mph]", ylab = "Distance [ft]")
points(speed[speed < 15], dist[speed < 15], pch = "s", col = "blue")
points(speed[speed >= 15], dist[speed >= 15], pch = "f", col = "green")
lines(lowess(cars), col = "red")
legend(5, 120, pch = c("s", "f"), col = c("blue", "green"),
legend = c("Slow", "Fast"))
title("Breaking distance of old cars")
111
Breaking distance of old cars
120
f
s Slow
f Fast
100 f
f f
80
s
f
Distance [ft]
f f
f f
60
s
f f f f
f f
s f f
f
40
s f f
s s f
s s f f f
s s s s f
s
20
s s s f
s s
s s
s s
0
5 10 15 20 25
Speed [mph]
detach(2)
112
Mean and Median of a Skewed Distribution
35
30 n xi
x=
25
i=1 n
Frequency
20
x^ = median(xi, i = 1, n)
15
10
5
0
0 2 4 6 8 10
x
1.0
sin
cos
0.5
f()
0.0
0.5
1.0
3 2 1 0 1 2 3
113
The par() function is used to access and modify the list of graphics parameters for the current
graphics device. See help on par() for more details.
To see a sample of point type available in R, type
example(pch)
library(manipulate)
manipulate(plot(1:x), x = slider(1, 100))
manipulate(plot(cars, xlim = c(0, x.max), type = type, ann = label,
col = col, pch = pch, cex = cex), x.max = slider(10, 25, step = 5,
initial = 25), type = picker(Points = "p", Line = "l", Step = "s"),
label = checkbox(TRUE, "Draw Labels"), col = picker(red = "red",
green = "green", yellow = "yellow"), pch = picker(`1` = 1, `2` = 2,
`3` = 3, `4` = 4, `5` = 5, `6` = 6, `7` = 7, `8` = 8, `9` = 9,
`10` = 10, `11` = 11, `12` = 12, `13` = 13, `14` = 14, `15` = 15,
`16` = 16, `17` = 17, `18` = 18, `19` = 19, `20` = 20, `21` = 21,
`22` = 22, `23` = 23, `24` = 24), cex = picker(`1` = 1, `2` = 2,
`3` = 3, `4` = 4, `5` = 5, `6` = 6, `7` = 7, `8` = 8, `9` = 9,
`10` = 10))
114
type = p, col=red, pch=19, cex=1 type = p, col=blue, pch=21, cex=0.5
120
120
100
100
80
80
dist
dist
60
60
40
40
20
20
0
0
5 10 15 20 25 5 10 15 20 25
speed speed
120
100
100
80
80
dist
dist
60
60
40
40
20
20
0
5 10 15 20 25 5 10 15 20 25
speed speed
115
6.4.2 R Colors
Thus far, we have frequently used numbers in plot to refer to a simple set of colors. There are
8 colors where 0:8 are white, black, red, green, blue, cyan, magenta, yellow and grey. If you
provide a number greater than 8, the colors are recycled. Therefore for plots where other or greater
numbers of colors are required, we need to access a larger palette of colors.
3
2
10
1
8
8
7
6
6
5
4
4
3
2
2
2 4 6 8 10 12
R has a large list of over 650 colors that R knows about. This list is held in the vector colors().
Have a look at this list, and maybe search for a set you are interested in.
colors()[1:10]
length(colors())
## [1] 657
116
## [3] "lightyellow" "lightyellow1"
## [5] "lightyellow2" "lightyellow3"
## [7] "lightyellow4" "yellow"
## [9] "yellow1" "yellow2"
## [11] "yellow3" "yellow4"
## [13] "yellowgreen"
R are has defined palettes of colors, which provide complementing or contrasting color sets.
For example look at the color palette rainbow.
example(rainbow)
Figure 6.1: There are 657 colors available using the colors() function
The complete versions of the above plots, along with a complete listing of colors, RGB numbers
117
or hexadecimal for each colors, the following script will generate a multi-page pdf document called
ColorChart.pdf which is a useful reference document on colors in R.
source("http://research.stowers-institute.org/efg/R/Color/Chart/ColorChart.
require(RColorBrewer)
118
YlOrRd
YlOrRd
YlOrBr
Set3 YlGnBu
YlOrBr YlGn
Reds
YlGnBu RdPu
Purples
Set2 YlGn PuRd
PuBuGn
Reds PuBu
OrRd
RdPu Oranges
Set1
Greys
Purples Greens
GnBu
PuRd BuPu
Pastel2 BuGn
PuBuGn Blues
PuBu Set3
Set2
Pastel1 Set1
OrRd
Pastel2
Oranges Pastel1
Paired
Paired Greys Dark2
Accent
Greens
Spectral
RdYlGn
GnBu
Dark2 RdYlBu
RdGy
BuPu
RdBu
PuOr
BuGn
PRGn
Accent
PiYG
Blues
BrBG
I use RColorBrewer to produce nicer colors in clustering heatmap. For example lets look at
results of wine-tasting of French red wines.
load the library ade4
119
##
## within
Vin_de_marque
Bordeaux_d_origine
Vin_de_table
Grand_Cru_classe
Cru_Bourgeois
excellent
good
boring
mediocre
120
Vin_de_marque
Bordeaux_d_origine
Vin_de_table
Grand_Cru_classe
excellent Cru_Bourgeois
good
boring
mediocre
121
6.5 Interacting with graphics
R also provides functions which allow users to extract or add information to a plot using a mouse
via locator() and verb+identify() functions respectively.
Waits for the user to select locations on the current plot using the left mouse button.
attach(women)
plot(height, weight)
identify(height, weight, women)
detach(2)
Allow the user to highlight any of the points (identify(x,y,label)) defined by x and
y (using the left mouse button) by plotting the corresponding component of labels nearby (or the
index number of the point if labels is absent).
Right mouse click, to stop.
Identify members in a hierarchical cluster analysis of distances between European cities
122
x
Height
0 1000 2000 3000 4000
Athens
Rome
Gibraltar
(x <- identify(hca))
Lisbon
Madrid
Stockholm
Copenhagen
Hamburg
123
Milan
Geneva
Lyons
eurodist
Barcelona
Marseilles
Brussels
Hook of Holland
Cherbourg
Calais
Paris
6.5.1 Exercise 8 - Plotting
Using the women dataset
1. Set the plot layout to be a 2 x 2 grid (ie 2 rows, 2 columns)
2. Draw weight on the Y axis and height on the X axis.
3. Switch the orientation, Draw weight on the X axis and height on the Y axis.
4. Drawing a new plot, set the pch (point type) to be a solid circle, and color them red. Add a
title study of Women to the plot
5. Drawing another plot, set the pch (point type) to be a solid square, Change the X axis label
to be Weight of Women and make the point size (using the parameter cex) larger to 1.5
124
6.6 Saving plots
6.6.1 Rstudio
In RStudio, there is a simple interface to export plots. Click on the Export button in the plot
window.
6.6.2 Devices
R can generate graphics (of varying levels of quality) on almost any type of display or printing
device. Before this can begin, however, R needs to be informed what type of device it is dealing
with. This is done by starting a device driver. The purpose of a device driver is to convert graphical
instructions from R (draw a line, for example) into a form that the particular device can under-
stand. Device drivers are started by calling a device driver function. There is one such function for
every device driver: type help(Devices) for a list of them all.
The most useful formats for saving R graphics:
postscript() For printing on PostScript printers, or creating PostScript graphics files.
pdf() Produces a PDF file, which can also be included into PDF files.
jpeg() Produces a bitmap JPEG file, best used for image plots.
125
When in doubt, I save files an postscript format (eps), as several journals request this format.
EPS files can be open directly in adobe illustrator or other vector editing graphics packages.
In R, to save the current image to file. Either use the file menu File -> Save as. Or use the
functions dev2bitmap, dev.copy2eps or dev.copy(device, file), where device can be one of png, jpeg
or pdf and file is your filename. For example:
To find out more about the image formats that can be saved in R, see the help on ?Devices.
If you wish to write an image directly to a file, without seeing the plot screen (called X11 or
Quartz depending on the operating system). Use the functions pdf(), postscript(), jpeg() with the
syntax:
pdf(file = "myplot.pdf")
plot(1:10, col = "blue", xlab = "X axis", ylab = "Y axis")
dev.off()
Remember it is very important to type dev.off in order to properly save the file
To list the current graphics devices that are open use dev.cur. When you have finished with a
device, be sure to terminate the device driver by issuing the command dev.off().
If you have open a device to write to for example pdf or png, dev.off will ensures that the
device finishes cleanly; for example in the case of hard-copy devices this ensures that every page
is completed and has been sent to the printer or file.
Example:
126
Chapter 7
Advanced Graphics
7.1.1 ggplots2
qplot is the basic plotting function in the ggplot2 package and is a convenient wrapper for creating a
number of different types of plots using a consistent calling scheme. See http://had.co.nz/
ggplot2/book/qplot.pdf for the chapter in the ggplot2 book which describes the usage of
qplot in detail.
A nice introductions to ggplots is written by its author Hadley Wickham and is available from
http://www.ceb-institute.org/bbs/wp-content/uploads/2011/09/handout_
ggplot2.pdf. The following examples are taking from that tutorial
Basic Quick Plot aka qplot in ggplots2
require("ggplot2")
data(mtcars)
head(mtcars)
127
## Mazda RX4 4
## Mazda RX4 Wag 4
## Datsun 710 1
## Hornet 4 Drive 1
## Hornet Sportabout 2
## Valiant 1
levels(mtcars$cyl)
## NULL
35
30
cyl
25 4
5
mpg
6
20 7
8
15
10
2 3 4 5
wt
128
35
30
25 factor(cyl)
4
mpg
6
20 8
15
10
2 3 4 5
wt
mtcars is a dataset from 1974 Motor Trend US magazine, and comprises fuel consumption and
10 aspects of automobile design and performance (eg miles/gallon, number of cylinders, displace-
ment, gross horsepower, weight, seconds to complete a quarter mile, etc) for 32 automobiles. In
the above plot we see a plot of weight x miles per gallon given the number of cylinders in the car.
In the above plot, we view cylinder by color, but it could also be by shape or size
129
35
30
25 factor(cyl)
4
mpg
6
20 8
15
10
2 3 4 5
wt
35
30
25 factor(cyl)
4
mpg
6
20 8
15
10
2 3 4 5
wt
The function mfrow and layout dont work with ggplots2, so here is a little script to make
layout of multiple plots using ggplots2 (acknowledgement to Stephen Turner). First assign each
ggplot2 plots to an object, and then use the arrange function to display two or more.
130
arrange <- function(..., nrow = NULL, ncol = NULL, as.table = FALSE) {
vp.layout <- function(x, y) viewport(layout.pos.row = x, layout.pos.col
dots <- list(...)
n <- length(dots)
if (is.null(nrow) & is.null(ncol)) {
nrow <- floor(n/2)
ncol <- ceiling(n/nrow)
}
if (is.null(nrow)) {
nrow <- ceiling(n/ncol)
}
if (is.null(ncol)) {
ncol <- ceiling(n/nrow)
}
## NOTE see n2mfrow in grDevices for possible alternative
grid.newpage()
pushViewport(viewport(layout = grid.layout(nrow, ncol)))
ii.p <- 1
for (ii.row in seq(1, nrow)) {
ii.table.row <- ii.row
if (as.table) {
ii.table.row <- nrow - ii.table.row + 1
}
for (ii.col in seq(1, ncol)) {
ii.table <- ii.p
if (ii.p > n)
break
print(dots[[ii.table]], vp = vp.layout(ii.table.row, ii.col))
ii.p <- ii.p + 1
}
}
}
131
35 35
30 30
25 factor(cyl) 25 factor(cyl)
4 4
mpg
mpg
6 6
20 8 20 8
15 15
10 10
2 3 4 5 2 3 4 5
wt wt
132
35
30
25
4
20
15
10
35
30
25
mpg
6
20
15
10
35
30
25
8
20
15
10
2 3 4 5
wt
More complex Facets to view cross-tabulated categories. For example you would expect a
strong interaction between cylinder and horsepower.
table(mtcars$cyl, mtcars$hp)
##
## 52 62 65 66 91 93 95 97 105 109 110 113 123 150 175 180 205 215
## 4 1 1 1 2 1 1 1 1 0 1 0 1 0 0 0 0 0 0
## 6 0 0 0 0 0 0 0 0 1 0 3 0 2 0 1 0 0 0
## 8 0 0 0 0 0 0 0 0 0 0 0 0 0 2 2 3 1 1
##
## 230 245 264 335
133
## 4 0 0 0 0
## 6 0 0 0 0
## 8 1 2 1 1
52 62 65 66 91 93 95 97 105 109 110 113 123 150 175 180 205 215 230 245 264 335
35
30
25
4
20
15
10
35
30
Miles per Gallon
25
6
20
15
10
35
30
25
8
20
15
10
2345 2345 2345 2345 2345 2345 2345 2345 2345 2345 2345 2345 2345 2345 2345 2345 2345 2345 2345 2345 2345 2345
Weight
In the scatterplot examples above, we implicitly used a point geom, the default when you supply
two arguments to qplot(). qplots can produce several other plots, if a different geom is defined
(modified from ggplots2: Elegant Graphics for Data Analysis, Chapter 3)
When given a single vector, the default geom is Histogram. Defining geom as density will
instead draw a density (smoothed histogram)
134
Plot Geom Other features
scatterplot point
bubblechart point size defined by variable
barchart bar
box-whisper boxplot
line line
135
5 3.0
2.5
4
2.0
3
count
count
1.5
2
1.0
1
0.5
0 0.0
2 3 4 5 2 3 4 5
wt wt
0.5 1.0
0.4 0.8
factor(cyl)
0.3 0.6
density
density
4
6
0.2 0.4 8
0.1 0.2
0.0 0.0
2 3 4 5 2 3 4 5
wt wt
136
arrange(p1, p2, p3, p4, nrow = 2, ncol = 2)
35 35
30
30
25
25
mpg
mpg
20
20
15
15
10
10 5
2 3 4 5 2 3 4 5
wt wt
30 0
30
25
20 factor(cyl) 5 factor(cyl)
15 1
factor(1)
10 4 4
count
5 25
0
6 6
8 10 8
20
15
factor(1) count
7.1.2 lattice
Lattice plots allow the use of the layout on the page to reflect meaningful aspects of data structure.
They offer abilities similar to those in the S-PLUS trellis library.
The lattice package sits on top of the grid package. To use lattice graphics, both these packages
must be installed. Providing it is installed, the grid package will be loaded automatically when
lattice is loaded.
Resources for lattice
137
Book on Lattice http://lmdvr.r-forge.r-project.org/figures/figures.
html
See examples at http://cm.bell-labs.com/cm/ms/departments/sia/project/
trellis/display.examples.html
To get on help on lattice functions, use help just like you would do for any package help(package
= lattice)
library(lattice)
xyplot(weight Time | Diet, data = ChickWeight) # Simple use of xyplot
138
0 5 10 15 20
3 4
300
200
100
weight
1 2
300
200
100
0 5 10 15 20
Time
Here is the statement used to get a figure with the observations for the same Chick connected
via lines.
139
0 5 10 15 20
3 4
300
200
100
weight
1 2
300
200
100
0 5 10 15 20
Time
This function shows the defaults for the graphical display of Trellis displays
show.settings()
An incomplete list of lattice Functions
splom( data.frame) # Scatterplot matrix
bwplot(factor numeric , . .) # Box and whisker plot
dotplot(factor numeric , . .) # 1-dim. Display
stripplot(factor numeric , . .) # 1-dim. Display
barchart(character numeric,...)
histogram( numeric, ...) # Histogram
densityplot( numeric, ...) # Smoothed version of histogram
qqmath(numeric numeric, ...) # QQ plot
splom( dataframe, ...) # Scatterplot matrix
parallelplot( dataframe, ...) # Parallel coordinate plots
In each instance, conditioning variables can be added.
Examples:
x <- 1:10
y <- 1:10
g <- factor(1:10)
140
barchart(y g | 1)
1
10
6
y
1 2 3 4 5 6 7 8 9 10
bwplot(yy gg | 1)
141
1
1.0
0.5
yy
0.0
0.5
1.0
1 2
densityplot(yy | 1)
0.4
0.3
Density
0.2
0.1
0.0
2 1 0 1 2
yy
142
histogram(yy | 1)
1
25
20
Percent of Total
15
10
yy
qqmath(yy | 1)
1
1.0
0.5
yy
0.0
0.5
1.0
2 1 0 1 2
qnorm
143
xyplot(xx yy | 1)
1
1.0
0.5
xx
0.0
0.5
1.0
yy
144
parallelplot(data.frame(x = xx[1:10], y = yy[1:10]) | 1)
1
y
Min Max
145
1
3.0
2.5
2.0
yyy
1.5
1.0
0.5
0.0
xxx
1
3.0
2.5
2.0
yyy
1.5
1.0
0.5
0.0
xxx
146
cloud(zzz xxx + yyy | 1, zlab = NULL, zoom = 0.9, par.settings = list(box.
yyy
xxx
yyy
xxx
147
Note whilst, lattice plots are highly customizable. Note: the base graphics settings; in partic-
ular, par() settings usually have no effect on lattice plots. Use trellis.par.get() and trellis.par.set()
to change default plot parameters.
148
7.2 GoogleVis and GoogleMaps visualization
There are multiple visualization tools available within the googleVis library. These include the
Hans Rosling type bubble plots. See http://code.google.com/apis/visualization/
documentation/gallery/motionchart.html
See http://blog.revolutionanalytics.com/graphics/ for some examples of
R code
# install.packages('googleVis')
library(googleVis)
GoogleVis also provide support for Maps and spatial visualization of trends.
149
Figure 7.1: Results of Fruits data using GoogleViz gvisMotionChart
This is actually an interactive animated html file
##
## Population
plot(states.Inc)
150
Figure 7.2: This is actually an interactive html file, if you hover over a state you see its information
require(ggmap)
# longtitude, latitude from google maps api
HSPH <- geocode("Harvard School of Public Health")
mapdist("Harvard School of Public Health", "New York", mode = "walking")
151
The qmap has a layered appear to maps similar to qplot in ggplot2. It allows you to extract a
map from google Maps or the other map distribution, and layer on data. A in-depth review of the
functions in ggmap is beyond the scope of this course, but I am happy to demo these should there
be sufficient interest.
There are online demonstrations on my website
http://bcb.dfci.harvard.edu/aedin/courses/R/CDC/maps.html
and of worldwide species diversity http://vijaybarve.wordpress.com/tag/rgbif/
and Napa valley vineyard http://blog.revolutionanalytics.com/2012/07/making-beaut
html
152
7.3 Graph theory and Network visualization using R packages network and
igraph
# install.packages(network)
library(network)
153
vertex.sides = 3 + apply(flo, 2, sum), vertex.col = 2 + (network.vertex.
"Medici"))
# install.packages('igraph')
library(igraph)
154
1
## -------------------------------------------------------- Given a
## list of PMIDs get their annotation Aedin, Dec 2011 To Run given
## pmids2tagcloud a list of pmids eg pmids=c(10521349, 10582678,
## 11004666, 11108479, 11108479, 11114790, 11156382, 11156382,
## 11156382, 11165872) pmids2tagcloud(pmids)
## ---------------------------------------------------------
getPMIDAnnot <- function(pmidlist) {
require(annotate)
require(XML)
print("Using annotate and XML to get info on each PMID")
pubmedRes <- xmlRoot(pubmed(pmidlist))
numAbst <- length(xmlChildren(pubmedRes))
absts <- list()
for (i in 1:numAbst) {
absts[[i]] <- buildPubMedAbst(pubmedRes[[i]])
}
155
PMIDInfo <- data.frame(matrix(NA, nrow = length(pmidlist)))
PMIDInfo$FirstAuthor <- unlist(lapply(absts, function(x) authors(x)[1]))
PMIDInfo$Journal <- unlist(lapply(absts, function(x) journal(x)[1]))
PMIDInfo$pubDate <- unlist(lapply(absts, function(x) pubDate(x)[1]))
PMIDInfo$articleTitle <- unlist(lapply(absts, function(x) articleTitle(x
PMIDInfo$abstText <- unlist(lapply(absts, function(x) abstText(x)[1]))
PMIDInfo$PubMedID <- unlist(lapply(absts, function(x) pmid(x)[1]))
rownames(PMIDInfo) <- PMIDInfo$PubMedID
PMIDInfo <- PMIDInfo[, -1]
# print(PMIDInfo[1:2,])
return(PMIDInfo)
}
156
pmids <- c(10521349, 10582678, 11004666, 11108479, 11108479,
11114790, 11156382, 11156382, 11156382, 11165872)
pmids2tagcloud(pmids)
leukemia
mnsodinduced
molecular
measured
diffuse cjun classification
tumor samples upregulated
cell
breast normal
signaling
genes cells microarray
dna
gene
cdna using et
blots consists
target
specific analysis
set
vivo cellular array
capture human ap1 approach
data classes found
pulmonary
laser
mnsod
gadd153
tissue
provide
results profile
class
grade
mcf+sod sets
determine rat antisense
revealed microdissection acute
expression
cancer phenotypic
development
PMIDS:10521349 10582678 11004666 11108479 11108479 11114790 11156382 11156382 11156382 11165872
157
2. For a discussion on different graph packages see
using Rgraphviz http://www2.warwick.ac.uk/fac/sci/moac/students/peter_
cock/r/rgraphviz/ or see the many examples on the Bioconductor website
a recent discussion online about the topic: http://stats.stackexchange.com/
questions/6155/graph-theory-analysis-and-visualization
R cytoscape http://db.systemsbiology.net:8080/cytoscape/RCytoscape/
vignette/RCytoscape.html
3. Additional demos available in the graphics package: demo(image), demo(persp) and exam-
ple(symbol).
The following web pages have many examples (and code) to produce different R plots. Browse
through the plots, see what you like and try some.
Basic plotting examples from Paul Murrell book, R Graphics http://www.stat.auckland.
ac.nz/paul/RGraphics/rgraphics.html
The homepage of the R package ggplot2 http://had.co.nz/ggplot2/. This pack-
age produces nice plots and can easily add color scale legend bars to plots
rggobi and ggplot2 run workshops in R graphics. Their course website provides examples
of basic and advanced plots, animated movies, lectures and R code to reproduce the plots at
http://lookingatdata.com/
The R Gallery wiki provides examples of R plots and code to reproduce these http:
//addictedtor.free.fr/graphiques/. Here is one random sample from this
website:
158
Chapter 8
159
See http://cran.r-project.org/src/contrib/Views/ for lists of more
R packages.
160
8.2 Basic Statistics
8.2.1 Continuous Data: t test
The t.test performs a one or two sample t test. To see the arguments of t.test, look at the help
documentation ?t.test
Arguments include alternative which is one of two.sided, less or greater, and var.equal
which is a logical (FALSE or TRUE) to indicate unequal or equal variance (default is unequal).
The input to t.test is one vector (one sample t test), two vectors or a formula (two sample t test). A
formula is given by y x, where the tilde operator specifies described by
One sample t test:
data(ChickWeight)
ChickWeight[1:2, ]
##
## One Sample t-test
##
## data: ChickWeight[, 1]
## t = 7.38, df = 577, p-value = 5.529e-13
## alternative hypothesis: true mean is not equal to 100
## 95 percent confidence interval:
## 116.0 127.6
## sample estimates:
## mean of x
## 121.8
##
##
## Welch Two Sample t-test
##
## data: ChickWeight$weight[ChickWeight$Diet == "1"] and ChickWeight$weight
## t = -2.638, df = 201.4, p-value = 0.008995
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
## -34.900 -5.042
## sample estimates:
161
## mean of x mean of y
## 102.6 122.6
##
##
## Welch Two Sample t-test
##
## data: weight by Diet
## t = -2.638, df = 201.4, p-value = 0.008995
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
## -34.900 -5.042
## sample estimates:
## mean in group 1 mean in group 2
## 102.6 122.6
##
pairwise.t.test(ChickWeight$weight, ChickWeight$Diet,
p.adjust.method="bonferroni")
##
## Pairwise comparisons using t tests with pooled SD
##
## data: ChickWeight$weight and ChickWeight$Diet
##
## 1 2 3
## 2 0.06838 - -
## 3 2.5e-06 0.14077 -
## 4 0.00026 0.95977 1.00000
##
## P value adjustment method: bonferroni
162
x <- rnorm(50, mean = c(rep(0, 25), rep(3, 25)))
p <- 2 * pnorm(sort(-abs(x)))
163
lmDiet <- lm(weight Diet, data = ChickWeight)
lmDiet
##
## Call:
## lm(formula = weight Diet, data = ChickWeight)
##
## Coefficients:
## (Intercept) Diet2 Diet3 Diet4
## 102.6 20.0 40.3 32.6
##
summary(lmDiet)
##
## Call:
## lm(formula = weight Diet, data = ChickWeight)
##
## Residuals:
## Min 1Q Median 3Q Max
## -104.0 -53.6 -13.6 40.4 230.1
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 102.65 4.67 21.96 < 2e-16 ***
## Diet2 19.97 7.87 2.54 0.011 *
## Diet3 40.30 7.87 5.12 4.1e-07 ***
## Diet4 32.62 7.91 4.12 4.3e-05 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 69.3 on 574 degrees of freedom
## Multiple R-squared: 0.0535,Adjusted R-squared: 0.0485
## F-statistic: 10.8 on 3 and 574 DF, p-value: 6.43e-07
##
anova(lmDiet)
164
In some statistical packages, the sum of the squares are labeled between groups and within
groups. Since lm and anova tables are used for a wide range of statistical models, the output from
R is different. The Between groups sum of the squares is labeled by the name of the factor group-
ings (Diet). The within sum of the squares is labeled Residuals. The aov function is a wrapper
which calls lm, but express the results these in the traditional language of the analysis of variance
rather than that of linear models.
## Call:
## aov(formula = weight Diet, data = ChickWeight)
##
## Terms:
## Diet Residuals
## Sum of Squares 155863 2758693
## Deg. of Freedom 3 574
##
## Residual standard error: 69.33
## Estimated effects may be unbalanced
##
## Call:
## lm(formula = weight Diet + Time, data = ChickWeight)
##
## Residuals:
## Min 1Q Median 3Q Max
## -136.8 -17.1 -2.6 15.0 141.8
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 10.924 3.361 3.25 0.0012 **
## Diet2 16.166 4.086 3.96 8.6e-05 ***
## Diet3 36.499 4.086 8.93 < 2e-16 ***
## Diet4 30.233 4.107 7.36 6.4e-13 ***
## Time 8.750 0.222 39.45 < 2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 36 on 573 degrees of freedom
165
## Multiple R-squared: 0.745,Adjusted R-squared: 0.744
## F-statistic: 419 on 4 and 573 DF, p-value: <2e-16
##
anova(lmDiet)
where
166
measure of association in a contingency table.
In R, first create a contingency table and then use the function assocstats to computes the Pear-
son chi-Squared test, the Likelihood Ratio, chi-Squared test, the phi coefficient, the contingency
coefficient, and Cramers V statistics.
## load library
library(vcd)
## load data
attach(Arthritis)
## [1] TRUE
print(levels(Arthritis$Treatment))
is.factor(Arthritis$Improved)
## [1] TRUE
print(levels(Arthritis$Improved))
##
## None Some Marked
## Placebo 29 7 7
## Treated 13 7 21
## compute statistics
res <- assocstats(tab)
print(res)
167
## X2 df P(> X2)
## Likelihood Ratio 13.530 2 0.0011536
## Pearson 13.055 2 0.0014626
##
## Phi-Coefficient : 0.394
## Contingency Coeff.: 0.367
## Cramer's V : 0.394
detach(Arthritis)
The structure of the res object can be printed using the str function
str(res)
## List of 5
## $ table : 'table' int [1:2, 1:3] 29 13 7 7 7 21
## ..- attr(*, "dimnames")=List of 2
## .. ..$ : chr [1:2] "Placebo" "Treated"
## .. ..$ : chr [1:3] "None" "Some" "Marked"
## $ chisq_tests: num [1:2, 1:3] 13.52981 13.05502 2 2 0.00115 ...
## ..- attr(*, "dimnames")=List of 2
## .. ..$ : chr [1:2] "Likelihood Ratio" "Pearson"
## .. ..$ : chr [1:3] "X2" "df" "P(> X2)"
## $ phi : num 0.394
## $ contingency: num 0.367
## $ cramer : num 0.394
## - attr(*, "class")= chr "assocstats"
You can easily access the various statistics from the res object
## X2 df P(> X2)
## 13.055020 2.000000 0.001463
## Cramer's V statistic
print(res$cramer)
## [1] 0.3942
If you want to compute the agreement between two classifications or raters, you can estimate
the coefficient which can have the following typical values
168
Kappa value magnitude of agreement
<0 no
0 - 0.2 small
0.2 - 0.4 fair
0.4 - 0.6 moderate
0.6 - 0.8 substantial
0.8 - 1 almost perfect
## value ASE
## Unweighted -0.04839 0.10073
## Weighted NaN 0.08598
In a practical situation, your Kappa coefficient needs to be over 0.6 to claim that your catego-
rization is valid. You may also want to report both the agreement (%)
## Agreement: 0.48%
Continuous Data
t.test
pairwise.t.test: pairwise comparisons
var.test: comparison of two variances.
lm(yx): linear regression analysis
lm(yf1): one-way analysis of variance
lm(yf1+f2): two-way analysis of variance (ANOVA), f1 and f2 are factors.
lm(yf1+x): analysis of co-variance
lm(yx1+x2+x3): multiple regression analysis
bartlett.test: Bartletts test of the null that the variances in each of the groups (samples)
are the same
Non-Parametric
169
wilcox.test: one- and two-sample Wilcoxon tests on vectors of data; the latter is also
known as Mann-Whitney test.
kruskal.test: non-parametric one-way analysis of variance.
friedman.test: non-parametric two-way analysis of variance.
Correlation
cor, cor.test: correlation and Correlation tests. Cor.test methods include kendall spear-
man or pearson.
Discrete response data
chisq.test: chi-squared contingency table tests, fisher.test exact test for small tables.
binom.test: binomial test
prop.test: prop.trend.test comparison of proportions.
glm(y x1+x2+x3, binomial): logistic regression
* As a complete aside and to continue stories of Irelands mathematicians and statisticians, which I started with the story of
George Boole, the first professor of mathematics of University College Cork. The t statistic was introduced by William Sealy
Gosset to monitoring the quality of brewing in the Guinness brewery in Dublin, Ireland. Guinnesss has an innovative policy of
recruiting the best graduates from Oxford and Cambridge to apply biochemistry and statistics to Guinness industrial processes.
Gosset published the t test in Biometrika in 1908, but published using the pen name Student.
170
8.3 Model formulae and model options
Most modeling done in a standard way
Data set is usually a single data frame object
Model is fitted using model fitting function
Form of the model specified by a formula
Resulting fitted model object can be interrogated, analyzed and modified
Basic output of the model fitting process is minimal. Details obtained via extractor functions.
171
8.3.1 Model formulae
We have already seen in several functions (boxplot, t.test, lm) that a simply function is defined by
y x. We will now discuss formulae in much more detail.
p
X
yi = j xij + i , i NID(0, 2 ), i = 1, . . . , n
j=0
y = X +
where y is the response vector, X is the model matrix or design matrix with columns x0 , x1 , . . . , xp .
NOTATION:
y, x, x0, x1, x2, ... - numeric variables
A, B, C, ... - factors
172
----------------------------------------------------------------------
y A*B + Error(C) - two factor model with interaction and error
strata determined by C
General form:
response op_1 term_1 + op_2 term_2 + ...
where
response - a vector or expression evaluating to a vector defining the
response variable
op_i - an operator, either '+' (inclusion of a term)
or '-' (exclusion of a term) in the model
term_i - either a vector or matrix expression,
or 1, a factor,
or a formula expression consisting of factors, vectors or
matrices connected by formula operators.
library(MASS)
help("cats")
str(cats)
173
8.3.3 Contrasts, model.matrix
We need to understand how model formulae specify the columns of the model matrix.
1. Continuous variables (simplest): each variable provides a column of the model matrix (and
the intercept will provide a column of ones if included in the model).
2. k-level factor A
The answer differs for unordered and ordered factors.
Unordered factors
k 1 columns are generated for the indicators of the second, third, . . . , up to kth levels
of the factor. (Implicit parameterization is to contrast the response at each level with
that at the first.)
Ordered factors
k 1 columns are the orthogonal polynomials on 1, . . . , k, omitting the constant term.
If the intercept is omitted in a model that contains a factor term, the first such term is encoded
into k columns giving the indicators for all the levels.
R default setting is:
contr.treatment(n = 3, base = 2)
## 1 3
## 1 1 0
## 2 0 0
## 3 0 1
contr.sum(n = 3)
## [,1] [,2]
## 1 1 0
## 2 0 1
## 3 -1 -1
174
8.4 Exercise 9
##
## Call:
## lm(formula = Hwt Sex, data = cats)
##
## Coefficients:
## (Intercept) SexM
## 9.20 2.12
##
##
## Call:
## lm(formula = Hwt Sex - 1, data = cats)
##
## Coefficients:
## SexF SexM
## 9.2 11.3
##
175
8.5 Output and extraction from fitted models
As mentioned earlier, the printed output of the model fit is minimal. However, the value of a fitted
model object is stored in an object. Information about the fitted model can be displayed, extracted
and plotted.
Extractor functions:
coef(obj) - regression coefficients
resid(obj) - residuals
fitted(obj) - fitted values
summary(obj) - analysis summary
predict(obj,newdata = ndat) - predict for new data
deviance(obj) - residual sum of squares
print(obj) - Print concise summary
plot(obj) - produce diagnostic plots
formula(obj) - extract the model formula
anova(obj1, obj2) - compare 2 models (one is a submodel with the outer mo
step(obj) - add, drop terms
update(obj, formula) - update a model with a new formula
Lets go back to the cats example. Now, we can extract more information about the fits.
attach(cats)
cats.lmBS <- lm(Hwt Bwt + Sex, data = cats)
coef(cats.lmBS)
plot(Bwt, Hwt)
lines(Bwt, fit.catsBS, col = "green", lwd = 2) # OR
# abline(cats.lmBS, col='green', lwd=2)
lines(Bwt[Sex == "F"], fit.catsBxS[Sex == "F"], col = "red")
lines(Bwt[Sex == "M"], fit.catsBxS[Sex == "M"], col = "blue")
legend(x = 2, y = 20, legend = c("LMBS", "lmBxS.female", "lmBxS.male"),
col = c("green", "red", "blue"), lwd = c(2, 1, 1))
176
20
LMBS
lmBxS.female
lmBxS.male
18
16
14
Hwt
12
10
8
6
## 1 2 3 4
## 7.441 11.754 16.067 20.379
detach(cats)
Some more useful, but non-standard, ways of extracting information from a model.
df.residual(obj) - residual degrees of freedom
names(obj) - gives names of the components in obj
177
names(summary(obj)) - gives names of the components in summary(obj)
How to get the residual variance of the fit? There are at least 2 ways. The first is the direct
calculation
# direct calculation
var.catsB <- deviance(cats.lmB)/df.residual(cats.lmB)
178
8.6 Exercise 10 :Multivariate linear regression
Read the data contained in the file lungs.csv from the course website into R. Fit a multi-
variate regression model (function lm) of pemax using all variables. Call the result lungFit.
Which terms appear to be significant (summary)?
What is the residual error of this model?
Which are the most and least significant variables in this model?
179
8.6.1 Residual plots, diagnostics
Important part of modeling - checking the model assumptions. Some easy to check and interpret
model diagnostics
1. Fit to the data (predictions) vs. raw data (observations)
2. Histogram of the residuals
3. Scatterplot of the residuals vs. fitted values
4. QQ-plot of the residuals
attach(cats)
plot(Bwt, Hwt, main = "Model fit")
abline(cats.lmB, col = "green", lwd = 2)
Residual histogram
180
qqnorm(resid(cats.lmB))
qqline(resid(cats.lmB))
detach()
Figure 8.1: Plots of the fitted Model,Residuals vs. fitted values and QQ-PLots of the residuals
QQ-plot of residuals
181
Default plots available in plot of class lm
4
144 144
3
4
135
135
Standardized residuals
71
71
2
2
Residuals
1
0
0
1
2
31
2
140 31
4
140
8 10 12 14 2 1 0 1 2
Fitted values Theoretical Quantiles
ScaleLocation Cook's distance
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35
144
144
135 140
1.5
31 71
Standardized residuals
Cook's distance
1.0
140
0.5
135
136
142
0.0
182
The sums of squares shown are the decrease in the residual sums of squares resulting from an
inclusion of that term in the model at that place in the sequence. Only for orthogonal experiments
will the order of inclusion be inconsequential.
anova(cats.lmB, cats.lmBS)
anova(cats.lmB, cats.lmBxS)
The update() function allows a model to be fitted that differs from one previously fitted usually
by just a few additional or removed terms.
Syntax:
new.model <- update(old.model, new.formula)
Special name in the new formula - a period . - can be used to stand for corresponding part of the
old model formula.
Example:
Data set mtcars, fuel consumption and 10 aspects of automobile design and performance for
32 automobiles.
183
help("mtcars")
cars.lm <- lm(mpg hp + wt, data = mtcars)
cars.lm2 <- update(cars.lm, . . + disp)
# cars.lms <- update(cars.lm2, sqrt(.) .)
addterm Fits all models that differ from the current model by adding a single term from those
supplied, maintaining marginality
stepAIC(model.small,
scope=list(upper=model.big, lower= 1),
test='F') # for linear models
184
Example
# Stepwise selection
cars.step <- stepAIC(cars.some, scope=list(lower = wt))
8.7 Cross-validation
Another approach, developed by the machine Learning community, is cross-validation. The idea
is to sequentially divide the dataset in training and test sets to fit and assess the performance of the
model, respectively. This approach enables to use all the observations both for training and testing
the prediction model.
Here is an example of a 10-fold cross-validation on the mtcars dataset where we compare the
model with one variable (wt) and all the variables to predict mpg. Once the root mean squared
error (RMSE) is computed for each fold, a paired Wilcoxon Rank Sum test is used to compare the
performance of the small and big models.
nfold <- 10
## nr is the number of observations
nr <- nrow(mtcars)
## nfold is the number of folds in the cross-validation
if(nfold > 1) k <- floor(nr/nfold) else {
k <- 1
nfold <- nr
}
smpl <- sample(nr)
mse.big <- mse.small <- NULL
for (i in 1:nfold) {
if (i == nfold) s.ix <- smpl[c(((i - 1) * k + 1):nr)] else s.ix <- s
## fit the model
185
mm.big <- lm(mpg ., data=mtcars[-s.ix, , drop=FALSE])
mm.small <- lm(mpg wt, data=mtcars[-s.ix, , drop=FALSE])
## assess the performance of the model
pp.big <- predict(object=mm.big, newdata=mtcars[s.ix, !is.element(colnames(m
pp.small <- predict(object=mm.small, newdata=mtcars[s.ix, !is.element(colnam
## compute mean squared error (MSE)
mse.big <- c(mse.big, sqrt(mean((mtcars[s.ix, "mpg"] - pp.big)2)))
mse.small <- c(mse.small, sqrt(mean((mtcars[s.ix, "mpg"] - pp.small)2)))
}
names(mse.big) <- names(mse.small) <- paste("fold", 1:nfold, sep=".")
## compare the performance of the big and small models using a Wilcoxon Rank
wilcox.test(mse.big, mse.small, paired=TRUE, alternative="less")
##
## Wilcoxon signed rank test
##
## data: mse.big and mse.small
## V = 35, p-value = 0.7842
## alternative hypothesis: true location shift is less than 0
##
As can be seen, there is not enough evidence in the dataset to claim that the big prediction
model outperforms the small one (p-value > 0.05). You can easily change the number of folds
in the cross-validation by setting the variable nfold to another value, nf old = 1 for leave-one-out
cross-validation.
186
8.8 Statistical models
Will talk today about 3 classes of statistical models: linear regression, generalized linear models
(e.g. logistic and Poisson regression), and survival models.
attach(ChickWeight)
time.wgt <- tapply(weight, Time, var)
time.wgt.rep <- as.numeric(time.wgt[match(Time, as.numeric(names(time.wgt)))
detach(2)
Chick.anl <- data.frame(ChickWeight, time.wgt.rep = time.wgt.rep)
chick.lm.wgt <- lm(weight Time, data = Chick.anl, weight = 1/time.wgt.rep)
chick.lm.T0 <- lm(weight Time, data = Chick.anl, subset = (Time ==
0))
187
8.8.2 Generalized linear modeling
One generalization of multiple linear regression.
Response, y, predictor variables x1 , x2 , . . . , xp
The distribution of Y depends on the xs through a single linear function, the linear predic-
tor
= 1 x1 + 2 x2 + . . . + p xp (8.1)
with xi having no influence on y if and only if i = 0
There may be an unknown scale (or variance) parameter to estimate as well
The mean, , is a smooth invertible function of the linear predictor
The class of generalized linear models handled by facilities supplied in R includes Gaussian, bino-
mial, Poisson, inverse Gaussian and gamma response distributions.
Families of distributions and links
Distribution Link
------------- -------------------------------
binomial logit, probit, log, cloglog
gaussian identity, log, inverse
Gamma identity, inverse, log
inverse.gaussian 1/mu2, identity, inverse, log
poisson identity, log, sqrt
The R function to fit a generalized linear model is glm() which uses the form
fitted.model <- glm(formula, family=family.generator, data=data.frame)
The only difference from lm() is the family.generator, which is the instrument by which the
family is described. It is the name of a function that generates a list of functions and expressions
that together define and control the model and estimation process.
We will concentrate on the binomial family with the logit link or as you probably know it
logistic regression,.
188
Logistic regression
To fit a binomial model using glm() there are three possibilities for the response:
1. If the response is a vector it is assumed to hold binary data, and so must be a 0/1 vector.
2. If the response is a two-column matrix it is assumed that the first column holds the number
of successes for the trial and the second holds the number of failures.
3. If the response is a factor, its first level is taken as failure (0) and all other levels as success
(1).
Syntax:
glm(y x, family=binomial(link=logit), data = data.frame)
Link is optional, since the default link is logit. Necessary, if another link is desired, e.g. probit.
Example of logistic regression using data set esophagus
Data from a case-control study of (o)esophageal cancer in Ile-et-Vilaine, France containing records
for 88 age/alcohol/tobacco combinations with the 3 covariates grouped into 6, 4 and 4 groups
respectively.
summary(esoph)
189
eso.base <- update(eso.age, . . + tobgp + alcgp)
## Start: AIC=298.6
## cbind(ncases, ncontrols) agegp
##
## Df Deviance AIC LRT Pr(Chi)
## + alcgp 3 64.6 230 74.5 4.5e-16 ***
## + tobgp 3 120.0 286 19.1 0.00026 ***
## <none> 139.1 299
## - agegp 5 227.2 377 88.1 < 2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Step: AIC=230.1
## cbind(ncases, ncontrols) agegp + alcgp
##
## Df Deviance AIC LRT Pr(Chi)
## + tobgp 3 54.0 226 10.6 0.014 *
## <none> 64.6 230
## - agegp 5 138.8 294 74.2 1.4e-14 ***
## - alcgp 3 139.1 299 74.5 4.5e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Step: AIC=225.4
## cbind(ncases, ncontrols) agegp + alcgp + tobgp
##
## Df Deviance AIC LRT Pr(Chi)
## <none> 54.0 226
## - tobgp 3 64.6 230 10.6 0.014 *
## - alcgp 3 120.0 286 66.1 3.0e-14 ***
## - agegp 5 131.5 293 77.5 2.8e-15 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
190
Prediction and residuals Plot the fitted values for age effect
attach(esoph)
eso.pred.age <- predict.glm(eso.base, data.frame(agegp = agegp,
tobgp = rep("30+", 88), alcgp = rep("40-79", 88)), type = "response")
plot(eso.pred.age agegp, ylab = "Predicted Age", xlab = "Age group (True a
0.4
0.3
Predicted Age
0.2
0.1
detach(esoph)
4 types of residuals can be requested for the glm() models: deviance, working, Pearson, re-
sponse
191
2
Different types of residuals
15
1
10
deviance
working
0
5
1
0
0 20 40 60 80 0 20 40 60 80
Index Index
4
0.4
3
0.2
2
response
pearson
1
0.0
0
0.2
1
0 20 40 60 80 0 20 40 60 80
Index Index
par(opar)
192
8.9 Survival modeling
Survival Analysis is a class of statistical methods for studying the occurrence and timing of events.
These methods are most often applied to the study of deaths but can also handle different kinds of
events, including the onset of disease and equipment failure for instance. For instance a disease
consists of a transition from an healthy state to a diseased state. Moreover, the timing of the event
is also considered for analysis.
Survival data have a common feature, namely censoring, that is difficult to handle with con-
ventional statistical methods. Consider the following example, which illustrates the problem of
censoring. A sample of breast cancer patients were followed during 10 years after diagnosis. The
event of interest was the appearance of a distant metastasis (a tumor initiated from the primary
breast tumor cells and that is located in another organ). The aim was to determine how the occur-
rence and timing of distant metastasis appearance depended on several variables.
The vast majority of the he functions we need to do survival analysis are in the package sur-
vival. Check if the package survival is already loaded into your work space, if it isnt load the
library survival
search()
library(survival)
We will work with the data set leukemia containing times of death or censoring in patients
with Acute Myelogenous Leukemia. The survival data are usually stored in a Surv object that is a
one-column matrix containing the survival times and events/censoring.
data(leukemia)
head(leukemia)
193
## time status x
## 1 9 1 Maintained
## 2 13 1 Maintained
## 3 13 0 Maintained
## 4 18 1 Maintained
## 5 23 1 Maintained
## 6 28 0 Maintained
`?`(Surv)
mysurv <- Surv(leukemia$time, leukemia$status)
head(mysurv)
Several methods for survival analysis are implemented in R, mainly in the survival package:
Surv - creates a survival object used as a response variable in a model formula,
e.g. Surv(time, status)
survfit - computes an estimate of a survival curve for censored data using the Kaplan-Meier
method, e.g. survfit(Surv(time, status) group)
survdiff - Tests if there is a difference between two or more survival curves,
e.g. survdiff(Surv(time, status) group)
survreg - regression for a parametric survival model with special case, the accelerated failure
models that use a log transformation of the response.
coxph - fits a Cox proportional hazards regression model
194
leuk.km <- survfit(Surv(time, status) x, data = leukemia)
plot(leuk.km, lty = 1, col = c("darkblue", "darkred"))
legend(100, 1, legend = c("Maintain", "Non-main"), lty = 1:2,
col = c("darkblue", "darkred"))
1.0
Maintain
Nonmain
0.8
0.6
0.4
0.2
0.0
0 50 100 150
195
## 48 2 1 0.184 0.1535 0.0117 0.525
##
## x=Nonmaintained
## time n.risk n.event survival std.err lower 95% CI upper 95% CI
## 5 12 2 0.8333 0.1076 0.48171 0.956
## 8 10 2 0.6667 0.1361 0.33702 0.860
## 12 8 1 0.5833 0.1423 0.27014 0.801
## 23 6 1 0.4861 0.1481 0.19188 0.730
## 27 5 1 0.3889 0.1470 0.12627 0.650
## 30 4 1 0.2917 0.1387 0.07240 0.561
## 33 3 1 0.1944 0.1219 0.03120 0.461
## 43 2 1 0.0972 0.0919 0.00575 0.349
## 45 1 1 0.0000 NaN NA NA
##
196
1.0
Maintain
Nonmain
0.8
0.6
0.4
0.2
0.0
0 50 100 150
## Call:
## survdiff(formula = Surv(time, status) x, data = leukemia)
##
## N Observed Expected (O-E)2/E (O-E)2/V
## x=Maintained 11 7 10.69 1.27 3.4
## x=Nonmaintained 12 11 7.31 1.86 3.4
##
## Chisq= 3.4 on 1 degrees of freedom, p= 0.0653
197
8.9.3 Cox proportional hazards model
The (semi-parametric) Cox regression model refers to the method first proposed in 1972 by the
British statistician Cox in his seminal paper Regression Models and Life Tables. It is difficult to
exaggerate the impact of this paper. In the 1992 Science Citation Index, it was cited over 800 times,
making it the most highly cited journal article in the entire literature of statistics. In fact, Garfield
reported that its cumulative citation count placed it among the top 100 papers in all branches of
science.
This enormous popularity can be explained by the fact that, unlike the parametric methods,
Coxs method does not require the selection of some particular probability distribution to represent
survival times. For this reason, the method is called semi-parametric. Cox made two significant
innovations. First, he proposed a model that is often referred to as the proportional hazards model.
Second, he proposed a new estimation method that was later named maximum partial likelihood.
The term Cox regression refers to the combination of the model and the estimation method
Here is an example of Cox regression estimating the benefit of maintaining chemotherapy of
with respect to the survival of the patients.
## Call:
## coxph(formula = Surv(time, status) x, data = leukemia)
##
## n= 23, number of events= 18
##
## coef exp(coef) se(coef) z Pr(>|z|)
## xNonmaintained 0.916 2.498 0.512 1.79 0.074 .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## exp(coef) exp(-coef) lower .95 upper .95
## xNonmaintained 2.5 0.4 0.916 6.81
##
## Concordance= 0.619 (se = 0.073 )
## Rsquare= 0.137 (max possible= 0.976 )
## Likelihood ratio test= 3.38 on 1 df, p=0.0658
## Wald test = 3.2 on 1 df, p=0.0737
## Score (logrank) test = 3.42 on 1 df, p=0.0645
##
It is not trivial to estimate the relevance of a variable with survival. If this variable is categori-
cal, you can draw the survival curves and statistically compare them. If the variable under interest
is continuous you can arbitrarily discretize it (not advisable) or use many existing performance cri-
teria published so far for survival analysis: hazard ration (see coxph), D.index, concordance.index,
198
time-dependent ROC curve, Brier score,. . . The survcomp package contains functions to estimate
these criteria.
199
8.10 Exercise 11: Survival Analysis
Use the colon dataset from the library survival
draw the Kaplan-Meier survival curves for the three group of patients encode by rx.
Use different colors for the curves and plot the lines twice as thick as the default size (param-
eter lwd).
Which color encodes which group? Add a legend to the plot to make this clear.
Generate a PDF output of the plot and put it in the website dropbox along with your code.
Test if the different patient group have significantly different outcome?
200
Chapter 9
data(mtcars)
df <- mtcars
dim(df)
## [1] 32 11
View data
View(df)
head(df)
tail(df)
201
## Ford Pantera L 15.8 8 351.0 264 4.22 3.170 14.5 0 1 5 4
## Ferrari Dino 19.7 6 145.0 175 3.62 2.770 15.5 0 1 5 6
## Maserati Bora 15.0 8 301.0 335 3.54 3.570 14.6 0 1 5 8
## Volvo 142E 21.4 4 121.0 109 4.11 2.780 18.6 1 1 4 2
str(df)
Basic Summary
summary(df)
202
library(Hmisc)
describe(df)
## df
##
## 11 Variables 32 Observations
## --------------------------------------------------------------------
## mpg
## n missing unique Mean .05 .10 .25 .50
## 32 0 25 20.09 12.00 14.34 15.43 19.20
## .75 .90 .95
## 22.80 30.09 31.30
##
## lowest : 10.4 13.3 14.3 14.7 15.0, highest: 26.0 27.3 30.4 32.4 33.9
## --------------------------------------------------------------------
## cyl
## n missing unique Mean
## 32 0 3 6.188
##
## 4 (11, 34%), 6 (7, 22%), 8 (14, 44%)
## --------------------------------------------------------------------
## disp
## n missing unique Mean .05 .10 .25 .50
## 32 0 27 230.7 77.35 80.61 120.83 196.30
## .75 .90 .95
## 326.00 396.00 449.00
##
## lowest : 71.1 75.7 78.7 79.0 95.1
## highest: 360.0 400.0 440.0 460.0 472.0
## --------------------------------------------------------------------
## hp
## n missing unique Mean .05 .10 .25 .50
## 32 0 22 146.7 63.65 66.00 96.50 123.00
## .75 .90 .95
## 180.00 243.50 253.55
##
## lowest : 52 62 65 66 91, highest: 215 230 245 264 335
## --------------------------------------------------------------------
## drat
## n missing unique Mean .05 .10 .25 .50
## 32 0 22 3.597 2.853 3.007 3.080 3.695
## .75 .90 .95
## 3.920 4.209 4.314
##
## lowest : 2.76 2.93 3.00 3.07 3.08, highest: 4.08 4.11 4.22 4.43 4.93
## --------------------------------------------------------------------
203
## wt
## n missing unique Mean .05 .10 .25 .50
## 32 0 29 3.217 1.736 1.956 2.581 3.325
## .75 .90 .95
## 3.610 4.048 5.293
##
## lowest : 1.513 1.615 1.835 1.935 2.140
## highest: 3.845 4.070 5.250 5.345 5.424
## --------------------------------------------------------------------
## qsec
## n missing unique Mean .05 .10 .25 .50
## 32 0 30 17.85 15.05 15.53 16.89 17.71
## .75 .90 .95
## 18.90 19.99 20.10
##
## lowest : 14.50 14.60 15.41 15.50 15.84
## highest: 19.90 20.00 20.01 20.22 22.90
## --------------------------------------------------------------------
## vs
## n missing unique Sum Mean
## 32 0 2 14 0.4375
## --------------------------------------------------------------------
## am
## n missing unique Sum Mean
## 32 0 2 13 0.4062
## --------------------------------------------------------------------
## gear
## n missing unique Mean
## 32 0 3 3.688
##
## 3 (15, 47%), 4 (12, 38%), 5 (5, 16%)
## --------------------------------------------------------------------
## carb
## n missing unique Mean
## 32 0 6 2.812
##
## 1 2 3 4 6 8
## Frequency 7 10 3 10 1 1
## % 22 31 9 31 3 3
## --------------------------------------------------------------------
204
table(df$cyl)
##
## 4 6 8
## 11 7 14
table(df$cyl, df$gear)
##
## 3 4 5
## 4 1 8 2
## 6 2 4 1
## 8 12 0 2
## , , = 0
##
##
## 3 4 5
## 4 1 2 0
## 6 2 2 0
## 8 12 0 0
##
## , , = 1
##
##
## 3 4 5
## 4 0 6 2
## 6 0 2 1
## 8 0 0 2
##
## gear
## 3 4 5
## 112 56 30
## , , vs = 0
##
## am
205
## gear 0 1
## 3 96 0
## 4 0 12
## 5 0 26
##
## , , vs = 1
##
## am
## gear 0 1
## 3 16 0
## 4 20 24
## 5 0 4
##
`?`(ftable)
ftable(df$cyl, df$vs, df$am, df$gear, row.vars = c(2, 4),
dnn = c("Cylinders", "V/S", "Transmission", "Gears"))
## Cylinders 4 6 8
## Transmission 0 1 0 1 0 1
## V/S Gears
## 0 3 0 0 0 0 12 0
## 4 0 0 0 2 0 0
## 5 0 1 0 1 0 2
## 1 3 1 0 2 0 0 0
## 4 2 6 2 0 0 0
## 5 0 1 0 0 0 0
## Cylinders 4 6 8
## Gears 3 4 5 3 4 5 3 4 5
## V/S Transmission
## 0 0 0 0 0 0 0 0 12 0 0
## 1 0 0 1 0 2 1 0 0 2
## 1 0 1 2 0 2 2 0 0 0 0
## 1 0 6 1 0 0 0 0 0 0
206
library(gmodels)
CrossTable(df$cyl, df$gear, format = "SAS")
##
##
## Cell Contents
## |-------------------------|
## | N |
## | Chi-square contribution |
## | N / Row Total |
## | N / Col Total |
## | N / Table Total |
## |-------------------------|
##
##
## Total Observations in Table: 32
##
##
## | df$gear
## df$cyl | 3 | 4 | 5 | Row Total |
## -------------|-----------|-----------|-----------|-----------|
## 4 | 1 | 8 | 2 | 11 |
## | 3.350 | 3.640 | 0.046 | |
## | 0.091 | 0.727 | 0.182 | 0.344 |
## | 0.067 | 0.667 | 0.400 | |
## | 0.031 | 0.250 | 0.062 | |
## -------------|-----------|-----------|-----------|-----------|
## 6 | 2 | 4 | 1 | 7 |
## | 0.500 | 0.720 | 0.008 | |
## | 0.286 | 0.571 | 0.143 | 0.219 |
## | 0.133 | 0.333 | 0.200 | |
## | 0.062 | 0.125 | 0.031 | |
## -------------|-----------|-----------|-----------|-----------|
## 8 | 12 | 0 | 2 | 14 |
## | 4.505 | 5.250 | 0.016 | |
## | 0.857 | 0.000 | 0.143 | 0.438 |
## | 0.800 | 0.000 | 0.400 | |
## | 0.375 | 0.000 | 0.062 | |
## -------------|-----------|-----------|-----------|-----------|
## Column Total | 15 | 12 | 5 | 32 |
## | 0.469 | 0.375 | 0.156 | |
## -------------|-----------|-----------|-----------|-----------|
##
##
207
## Warning: Chi-squared approximation may be incorrect
##
##
## Cell Contents
## |-------------------------|
## | N |
## | Expected N |
## | Chi-square contribution |
## | N / Row Total |
## | N / Col Total |
## | N / Table Total |
## |-------------------------|
##
##
## Total Observations in Table: 32
##
##
## | df$gear
## df$cyl | 3 | 4 | 5 | Row Total |
## -------------|-----------|-----------|-----------|-----------|
## 4 | 1 | 8 | 2 | 11 |
## | 5.156 | 4.125 | 1.719 | |
## | 3.350 | 3.640 | 0.046 | |
## | 0.091 | 0.727 | 0.182 | 0.344 |
## | 0.067 | 0.667 | 0.400 | |
## | 0.031 | 0.250 | 0.062 | |
## -------------|-----------|-----------|-----------|-----------|
## 6 | 2 | 4 | 1 | 7 |
## | 3.281 | 2.625 | 1.094 | |
## | 0.500 | 0.720 | 0.008 | |
## | 0.286 | 0.571 | 0.143 | 0.219 |
## | 0.133 | 0.333 | 0.200 | |
## | 0.062 | 0.125 | 0.031 | |
## -------------|-----------|-----------|-----------|-----------|
## 8 | 12 | 0 | 2 | 14 |
## | 6.562 | 5.250 | 2.188 | |
## | 4.505 | 5.250 | 0.016 | |
## | 0.857 | 0.000 | 0.143 | 0.438 |
## | 0.800 | 0.000 | 0.400 | |
## | 0.375 | 0.000 | 0.062 | |
## -------------|-----------|-----------|-----------|-----------|
## Column Total | 15 | 12 | 5 | 32 |
## | 0.469 | 0.375 | 0.156 | |
## -------------|-----------|-----------|-----------|-----------|
##
208
##
## Statistics for All Table Factors
##
##
## Pearson's Chi-squared test
## ------------------------------------------------------------
## Chi2 = 18.04 d.f. = 4 p = 0.001214
##
##
##
library(gmodels)
CrossTable(df$cyl, df$gear, format = "SPSS")
##
## Cell Contents
## |-------------------------|
## | Count |
## | Chi-square contribution |
## | Row Percent |
## | Column Percent |
## | Total Percent |
## |-------------------------|
##
## Total Observations in Table: 32
##
## | df$gear
## df$cyl | 3 | 4 | 5 | Row Total |
## -------------|-----------|-----------|-----------|-----------|
## 4 | 1 | 8 | 2 | 11 |
## | 3.350 | 3.640 | 0.046 | |
## | 9.091% | 72.727% | 18.182% | 34.375% |
## | 6.667% | 66.667% | 40.000% | |
## | 3.125% | 25.000% | 6.250% | |
## -------------|-----------|-----------|-----------|-----------|
## 6 | 2 | 4 | 1 | 7 |
## | 0.500 | 0.720 | 0.008 | |
## | 28.571% | 57.143% | 14.286% | 21.875% |
## | 13.333% | 33.333% | 20.000% | |
## | 6.250% | 12.500% | 3.125% | |
## -------------|-----------|-----------|-----------|-----------|
## 8 | 12 | 0 | 2 | 14 |
## | 4.505 | 5.250 | 0.016 | |
## | 85.714% | 0.000% | 14.286% | 43.750% |
209
## | 80.000% | 0.000% | 40.000% | |
## | 37.500% | 0.000% | 6.250% | |
## -------------|-----------|-----------|-----------|-----------|
## Column Total | 15 | 12 | 5 | 32 |
## | 46.875% | 37.500% | 15.625% | |
## -------------|-----------|-----------|-----------|-----------|
##
##
##
## Cell Contents
## |-------------------------|
## | Count |
## | Expected Values |
## | Chi-square contribution |
## | Row Percent |
## | Column Percent |
## | Total Percent |
## |-------------------------|
##
## Total Observations in Table: 32
##
## | df$gear
## df$cyl | 3 | 4 | 5 | Row Total |
## -------------|-----------|-----------|-----------|-----------|
## 4 | 1 | 8 | 2 | 11 |
## | 5.156 | 4.125 | 1.719 | |
## | 3.350 | 3.640 | 0.046 | |
## | 9.091% | 72.727% | 18.182% | 34.375% |
## | 6.667% | 66.667% | 40.000% | |
## | 3.125% | 25.000% | 6.250% | |
## -------------|-----------|-----------|-----------|-----------|
## 6 | 2 | 4 | 1 | 7 |
## | 3.281 | 2.625 | 1.094 | |
## | 0.500 | 0.720 | 0.008 | |
## | 28.571% | 57.143% | 14.286% | 21.875% |
## | 13.333% | 33.333% | 20.000% | |
## | 6.250% | 12.500% | 3.125% | |
## -------------|-----------|-----------|-----------|-----------|
## 8 | 12 | 0 | 2 | 14 |
## | 6.562 | 5.250 | 2.188 | |
## | 4.505 | 5.250 | 0.016 | |
210
## | 85.714% | 0.000% | 14.286% | 43.750% |
## | 80.000% | 0.000% | 40.000% | |
## | 37.500% | 0.000% | 6.250% | |
## -------------|-----------|-----------|-----------|-----------|
## Column Total | 15 | 12 | 5 | 32 |
## | 46.875% | 37.500% | 15.625% | |
## -------------|-----------|-----------|-----------|-----------|
##
##
## Statistics for All Table Factors
##
##
## Pearson's Chi-squared test
## ------------------------------------------------------------
## Chi2 = 18.04 d.f. = 4 p = 0.001214
##
##
##
## Minimum expected frequency: 1.094
## Cells with Expected Frequency < 5: 6 of 9 (66.67%)
##
211
Chapter 10
Basic scatterplot
attach(df)
8
25
mpg
20
15
10
16 18 20 22
qsec
212
10.1 Hierarchical Cluster Analysis
Or calculate correlation and view on heatmap
heatmap(cor(df))
carb
hp
disp
cyl
wt
gear
am
drat
mpg
vs
qsec
qsec
vs
mpg
drat
am
gear
wt
cyl
disp
hp
carb
We can look at the US state fact and figure information in the package state, which contains a
matrix called state.x77 containing information on 50 US states (50 rows) on population, income,
Illiteracy, life expectancy, murder, high school graduation, number of days with frost, and area
(8 columns). The default clustering of this uses a rather ugly red-yellow color scheme which I
changed to a red/brown-blue.
require(RColorBrewer)
hmcol <- colorRampPalette(brewer.pal(10, "RdBu"))(500)
heatmap(t(state.x77), col = hmcol, scale = "row")
213
10.2
Count
0
4
0
2
Color Key
4
Row ZScore
and Histogram
Alaska
Texas
Montana
California
Colorado Alaska
Wyoming Texas
Montana
Oregon California
Colorado
New Mexico Wyoming
Nevada Oregon
Mexico
NewNevada
Arizona
West Virginia West Arizona
Virginia
Maine
Maine South
RhodeCarolina
Island
South Carolina Delaware
Rhode Island Massachusetts
214
New Jersey New Hampshire
North Dakota
Hawaii Washington
Connecticut Oklahoma
Missouri
Maryland South Dakota
Nebraska
Vermont Minnesota
Kansas
New Hampshire Utah
North Dakota NewIdahoYork
Washington Ohio
Pennsylvania
Oklahoma Indiana
Virginia
Missouri Kentucky
Tennessee
South Dakota Arkansas
Alabama
Nebraska Carolina
NorthLouisiana
Minnesota Mississippi
Kansas Georgia
Iowa
Utah Wisconsin
Florida
Idaho Michigan
Illinois
New York
Ohio
Pennsylvania
Indiana
Virginia
Area
Frost
Kentucky
Tennessee
Murder
Income
Arkansas
Life Exp
Illiteracy
HS Grad
Alabama
North Carolina
Population
Louisiana
Mississippi
Georgia
Iowa
Wisconsin
Florida
Michigan
Illinois
Area
Frost
Murder
Income
Life Exp
Illiteracy
HS Grad
heatmap.2(t(state.x77), col = hmcol, scale = "row", trace = "none")
Population
res <- prcomp(df)
screeplot(res)
res
15000
Variances
10000
5000
0
biplot(res)
215
Cadillac Fleetwood
Pontiac Firebird
0.2 Hornet 4 Drive
Merc 240D Lincoln Continental
Dodge Challenger
Valiant
Hornet
AMC Sportabout
Javelin
Honda Civic Chrysler Imperial
Merc 230 disp
Fiat X19
FiatCorolla
Toyota 128
Mazda
PorscheMazdaRX4 Wag
RX4
9142 qsec
0.0
mpg
gear
vs
cyl
am
wt
drat
carb
Toyota
DatsunCorona
710
Merc280C
Merc 280
Merc 450SL
Merc450SLC
450SE
Volvo 142E
hp
Lotus Europa
Duster 360
PC2
Camaro Z28
0.2
Ford Pantera L
Ferrari Dino
0.4
0.6
Maserati Bora
Or using dudi.pca in ade4. The R package has some very nicely graphicing tools for visualizing
the results of PCA
216
s.class(res$li, factor(mpg < 20, labels = c("<20", ">20")),
sub = "MPG of cars", col = c("green4", "orange"))
6 d = 0.5
qsec
5
4
vs
wt
3
disp
mpg
cyl
2
hp
1
drat
carb am
0
gear
Car features
d=2 d=2
4 <20
8 6 >20
217
Lets look at the doubs data in the ade4 package. This data set gives environmental variables,
fish species and spatial coordinates for 30 sites
require(ade4)
data(doubs)
lapply(doubs, head, 2)
## $env
## dfs alt slo flo pH har pho nit amm oxy bdo
## 1 3 934 6.176 84 79 45 1 20 0 122 27
## 2 22 932 3.434 100 80 40 2 20 10 103 19
##
## $fish
## Cogo Satr Phph Neba Thth Teso Chna Chto Lele Lece Baba Spbi Gogo
## 1 0 3 0 0 0 0 0 0 0 0 0 0 0
## 2 0 5 4 3 0 0 0 0 0 0 0 0 0
## Eslu Pefl Rham Legi Scer Cyca Titi Abbr Icme Acce Ruru Blbj Alal
## 1 0 0 0 0 0 0 0 0 0 0 0 0 0
## 2 0 0 0 0 0 0 0 0 0 0 0 0 0
## Anan
## 1 0
## 2 0
##
## $xy
## x y
## 1 88 7
## 2 94 14
##
## $species
## Scientific French English code
## 1 Cottus gobio chabot european bullhead Cogo
## 2 Salmo trutta fario truite fario brown trout Satr
##
218
d=1
15
29 30
14
16
13 17
Ax3 4 12 1819 22
3 7 2021 28
2 6 11 27
Ax1
10 5
1 9 26
Ax2
X axes
8
x1
24
23
Ax1
25
Ax2
Y axes
d = 0.2 d = 0.2
Eigenvalues oxy
Phph
x1 Satr Neba
pH
flo
Baba
Teso dfs
Thth
Cogo har
Spbi
Pefl
Chto
Lele Rham
Cyca
Titi
Anan
AbbrBlbj
LegiGogo
Icme
Eslu
Lece alt
Chna slo nit
Scer Acce
Ruru
pho
amm
Alal
bdo
Y Canonical weights X Canonical weights
219
Chapter 11
The aim of reproducible research is to provide code such that an experiement/analysis can be easily
repeated. The best approach is to create a document of results, in which the code to generate those
results are embedded.
For more options see help on spin (?spin). The markdown file can be easily convert to MSOffice
using pandoc. There is an example below.
Another function from Knitr called stitch() will insert an R script into a template to create a
simple report. Knitr supplies a template web page (html), markdown and latex file.
11.2 Sweave
Therefore there is broad support for Sweave. Most of the documentation in R is written in Sweave
files
Seaves weaves R (or S) code into latex. This document is written in Sweave, but had recently
been converted to knitr (more about how to do that later).
You can also write R code embedded in a html template file and process this using Sweave
In each case the native document format is written tex or html, but R code is put in chunks.
These chunks are marked by <<>> and @.
You can also put shorter snippets of R code in text using the function eg \Sexprs{a3+ will
evaluate a3 +
220
11.2.1 Converting an Rnw file in Sweave to Knitr
Whilst Sweave is still used widely, increasingly its being replaced by knitr which has more func-
tionality.
Its easy to convert a Sweave rnw file to a knitr rnw
knitr does not allow spaces in code chunks, so either remove or subsitute spaces for any other
character eg
knitr has many more option for formating code withint the document. Most of these are not
available in Sweave so these do not affect conversion. For example, Sweave has the results
options verbatim/hide/tex, knitr has options markup, hide, asis. Note knitr requires the
option be enclosed in single quotes.
```{r}
summary(cars)
```
221
```{r fig.width=7, fig.height=6}
plot(cars)
```
Then to create the markdown file with embedded R code and results, run the command knit.
222
Chapter 12
Solutions to Exercises
223
nrow(women)
## [1] 15
ncol(women)
## [1] 3
dim(women)
## [1] 15 3
summary(women)
colMeans(women)
colnames(women)
`?`(colnames)
sum(women$weight < 120)
## [1] 2
women[order(women$weight), ]
224
## 11 68 146 34
## 12 69 150 36
## 13 70 154 33
## 14 71 159 30
## 15 72 164 37
## [1] 65
## [1] 60
nrow(TG2)
## [1] 60
mean(TG$len)
## [1] 18.81
sd(TG$len)
## [1] 7.649
mean(TG2$len)
## [1] 18.81
sd(TG2$len)
## [1] 7.649
225
## Analysis of Variance Table
##
## Response: len
## Df Sum Sq Mean Sq F value Pr(>F)
## supp 1 205 205 11.4 0.0013 **
## dose 1 2224 2224 124.0 6.3e-16 ***
## Residuals 57 1023 18
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## [1] 16
ncol(women)
## [1] 1
colnames(women)
## [1] "X.html."
summary(women)
## X.html.
## </div> :2
## </body> :1
## </head> :1
## </html> :1
## </td></tr></table> :1
## <body style=background-color:#fff>:1
## (Other) :9
226
myVec <- c(LETTERS[1:20], seq(0, 200, 10))
myVec
## [1] "A" "B" "C" "D" "E" "F" "G" "H" "I" "J"
## [11] "K" "L" "M" "N" "O" "P" "Q" "R" "S" "T"
## [21] "0" "10" "20" "30" "40" "50" "60" "70" "80" "90"
## [31] "100" "110" "120" "130" "140" "150" "160" "170" "180" "190"
## [41] "200"
## [1] "T" "70" "80" "200" "160" "130" "Q" "M" "100" "120"
## [11] "180" "30" "S" "110" "50" "N" "I" "190" "C" "10"
## [21] "A" "F" "140" "P" "0" "20" "B" "G" "150" "J"
## [31] "40" "L" "60" "90" "H" "E" "170" "R" "K" "D"
## [41] "O"
## [1] "T" "70" "80" "200" "160" "130" "Q" "M" "100" "120"
## [1] 2
## [1] 4
## [1] 8
## [1] 16
## [1] 32
## [1] 64
## [1] 128
## [1] 256
## [1] 512
## [1] 1024
227
x <- 1
while (2x < 1000) {
print(2x)
x <- x + 1
}
## [1] 2
## [1] 4
## [1] 8
## [1] 16
## [1] 32
## [1] 64
## [1] 128
## [1] 256
## [1] 512
require(XML)
## Reads all the tables in the webpage into a list
worldPop <- readHTMLTable("http://en.wikipedia.org/wiki/World_population")
## [1] "list"
length(worldPop)
## [1] 19
## [1] "NULL"
## [2] "toc"
## [3] "NULL"
## [4] "World population milestones (USCB estimates)"
## [5] "The 10 countries with the largest total population:"
## [6] "10 most densely populated countries (with population above 1"
## [7] "Countries ranking highly in terms of both total population ("
## [8] "NULL"
228
## [9] "UN (medium variant 2010 revision) and US Census Burea"
## [10] "UN 2008 estimates and medium variant projections (in million"
## [11] "World historical and predicted populations (in millions)[101"
## [12] "World historical and predicted populations by percentage dis"
## [13] "Estimated world and regional populations at various dates (i"
## [14] "Starting at 500 million"
## [15] "Starting at 375 million"
## [16] "NULL"
## [17] "NULL"
## [18] "NULL"
## [19] "NULL"
To tidy up this tables, lets look at dates after 1750AD, so lets remove rows 1 to 14 as these
have only world population information. Also we will remove row 32 which is just column names
## Now lets check the structure of the table The data are factors,
## lets convert to characters as these are easier to edit
str(worldPop)
## chr [1:31, 1:9] "70,000 BC" "10,000 BC" "9000 BC" ...
## - attr(*, "dimnames")=List of 2
## ..$ : NULL
## ..$ : chr [1:9] "Year" "World" "Africa" "Asia" ...
229
## Remove rows 1-14, 32
worldPop <- worldPop[-c(1:14, 32), ]
## Lets get rid of 'Note 1' To get rid of the square brakets we
## have to put \\ before them
colnames(worldPop) <- sub("\\[Note 1\\]", "", colnames(worldPop))
str(worldPop)
# In what year did the population of Europe, Africa and Asia exceed
# 500 million?
for (i in c("Europe", "Asia", "Africa")) {
print(paste(i, min(worldPop$Year[worldPop[, i] > 500]), sep = " "))
}
230
## Create a empty Plot
plot(worldPop$Year, worldPop$Asia, xlab = "Year", ylab = "Population (millio
col = "red", type = "n", ylim = c(200, 4000))
for (i in 1:length(regions)) {
region <- regions[i]
print(region)
lines(worldPop$Year, worldPop[, region], col = i, type = "l")
}
## [1] "Africa"
## [1] "Asia"
## [1] "Europe"
## [1] "Latin America"
## [1] "Northern America"
## [1] "Oceania"
231
4000
Africa
Asia
Europe
Latin America
Northern America
Oceania
3000
Population (millions)
2000
1000
232
## Mean :18.8 Mean :1.17
## 3rd Qu.:25.3 3rd Qu.:2.00
## Max. :33.9 Max. :2.00
233
72
160
70
150
68
66
weight
height
140
64
130
62
60
120
58
58 60 62 64 66 68 70 72 120 130 140 150 160
height weight
Study of Women
72
72
70
70
height of women (inches)
68
68
66
66
height
64
64
62
62
60
60
58
58
120 130 140 150 160 120 130 140 150 160
weight Weight of Women (lbs)
234
12.9 Solution to Exercise 9
## (Intercept) SexM
## 1 1 0
## 2 1 0
## 3 1 0
## 4 1 0
## 5 1 0
## 6 1 0
## 7 1 0
## 8 1 0
## 9 1 0
## 10 1 0
## SexF SexM
## 1 1 0
## 2 1 0
## 3 1 0
## 4 1 0
## 5 1 0
## 6 1 0
## 7 1 0
## 8 1 0
## 9 1 0
## 10 1 0
235
lungFit <- lm(pemax ., data = data.lungs)
summary(lungFit)
##
## Call:
## lm(formula = pemax ., data = data.lungs)
##
## Residuals:
## Min 1Q Median 3Q Max
## -37.34 -11.53 1.08 13.39 33.41
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 176.058 225.891 0.78 0.45
## age -2.542 4.802 -0.53 0.60
## sex -3.737 15.460 -0.24 0.81
## height -0.446 0.903 -0.49 0.63
## weight 2.993 2.008 1.49 0.16
## bmp -1.745 1.155 -1.51 0.15
## fev1 1.081 1.081 1.00 0.33
## rv 0.197 0.196 1.00 0.33
## frc -0.308 0.492 -0.63 0.54
## tlc 0.189 0.500 0.38 0.71
##
## Residual standard error: 25.5 on 15 degrees of freedom
## Multiple R-squared: 0.637,Adjusted R-squared: 0.42
## F-statistic: 2.93 on 9 and 15 DF, p-value: 0.032
##
resid(lungFit)
## 1 2 3 4 5 6 7 8
## 10.031 -3.414 13.386 -11.532 18.691 -31.552 -11.480 20.034
## 9 10 11 12 13 14 15 16
## -20.307 -13.182 15.646 10.748 -3.664 -33.118 10.460 33.405
## 17 18 19 20 21 22 23 24
## 21.034 -3.002 12.096 1.081 -37.338 11.864 -4.332 -34.233
## 25
## 28.677
## most significant
sort(summary(lungFit)$coefficients[, 4], decreasing = FALSE)[1]
## bmp
## 0.1517
236
## least significant
sort(summary(lungFit)$coefficients[, 4], decreasing = TRUE)[1]
## sex
## 0.8123
library(survival)
head(colon)
237
1.0
0.8
0.6
0.4
0.2
0.0
238
1.0
0.8
0.6
0.4
0.2
Obs
Lev
0.0
Lev+5FU
pdf(file = "Surv.pdf")
plot(colonFit, col = 2:4, lwd = 2)
legend("bottomleft", legend = levels(colon$rx), fill = 2:4)
dev.off()
## pdf
## 2
239
## Call:
## survdiff(formula = Surv(time, status) rx, data = colon)
##
## N Observed Expected (O-E)2/E (O-E)2/V
## rx=Obs 630 345 299 7.01 10.40
## rx=Lev 620 333 295 4.93 7.26
## rx=Lev+5FU 608 242 326 21.61 33.54
##
## Chisq= 33.6 on 2 degrees of freedom, p= 4.99e-08
## Call:
## coxph(formula = Surv(time, status) rx, data = colon)
##
## n= 1858, number of events= 920
##
## coef exp(coef) se(coef) z Pr(>|z|)
## rxLev -0.0209 0.9793 0.0768 -0.27 0.79
## rxLev+5FU -0.4410 0.6434 0.0839 -5.26 1.5e-07 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## exp(coef) exp(-coef) lower .95 upper .95
## rxLev 0.979 1.02 0.842 1.138
## rxLev+5FU 0.643 1.55 0.546 0.758
##
## Concordance= 0.545 (se = 0.009 )
## Rsquare= 0.019 (max possible= 0.999 )
## Likelihood ratio test= 35.2 on 2 df, p=2.23e-08
## Wald test = 33.1 on 2 df, p=6.45e-08
## Score (logrank) test = 33.6 on 2 df, p=4.99e-08
##
240
12.12 SessionInfo
sessionInfo()
241